Wikipedia:Village pump (policy)

 Policy Technical Proposals Idea lab WMF Miscellaneous 

The policy section of the village pump is intended for discussions about already-proposed policies and guidelines, as well as changes to existing ones. Discussions often begin on other pages and are subsequently moved or referenced here to ensure greater visibility and broader participation.

See the list of frequently rejected or ignored proposals. Discussions are automatically archived after 7 days of inactivity. To keep this page's size accessible, discussions with more than about 100 comments should be split to a separate page.

RfC: Should templates containing article body content be allowed?

Under what circumstances should templates be allowed to contain article body text?

Option 1: Never (i.e. substitute out all templates in Category:Article shared content templates)

Option 2A (de facto status quo): When content is sufficiently similar between articles that the same wording may be used with only changes in parameters (e.g. {{Named storms intro}} being used for List of named storms (A) to List of named storms (Z))

Option 2B (closest to de jure status quo): Option 2A but only when there is no expectation that the content within the template will need to be edited

Sdkb talk 16:05, 22 October 2026 (UTC)

Background

This RfC seeks to resolve a discrepancy between the WP:Template namespace guideline, which historically advised that Templates should not normally be used to store article text, as this makes it more difficult to edit the content, and the longstanding status quo practice of having such prose templates (see the 54 entries and 6 subcategories of Category:Article shared content templates, which was created in 2008). Editors have disagreed over how to interpret "normally" in the guideline, particularly during a long-running dispute over the existence of the {{Rotten Tomatoes prose}} and {{Metacritic film prose}} templates, with a December 2023 RfC and June 2026 TfD reaching different conclusions. A recent attempt to adjust the guideline text did not garner significant engagement due to participant confusion. Editors suggested instead asking directly about the permissibility of prose templates, which has led to this discussion. Sdkb talk 16:05, 22 October 2026 (UTC)

Survey (shared content templates)

  • Option 2A. This RfC speaks to the heart of why templates exist. They exist because, when information is used in multiple places, we want to centralize it rather than copying it so that we can maintain, optimize, and standardize it in a single place. This principle (expressed in the software world as don't repeat yourself) is as fundamental to reducing the workload needed to make Wikipedia as the content scope limitations imposed by our notability requirements. It is obvious for templates like infoboxes — imagine what a disaster it would be if we tried to design an infobox from scratch for every article, debating which fields to include and how to order them — but it applies equally well to some instances of text that is reused between articles. This includes A-Z series of articles that should have the same introduction (e.g. {{List of post offices in Colorado header}}) as well as sections/paragraphs that may be reused with only a few changes that can be captured by parameters (e.g. {{Year article header}} and {{Rotten Tomatoes prose}}).
  • Prose templates with thousands of transclusions have been used for years, with good reason:
    1. Templated text is standardized, which is often good. Readers become accustomed to certain aspects of Wikipedia style over time, and when articles are similar, it makes it easier to navigate unfamiliar pages because they know what to look for.
    2. Templating allows for optimization. When a format is applied over hundreds of pages, it becomes worthwhile to refine small details that probably would never have been considered at the level of an individual page. Disputes can be resolved at centralized discussion rather than playing out diffusely on thousands of low-visibility pages.
    3. Templates can be improved over time. When circumstances or best practices change (such as Rotten Tomatoes deprecating the average score metric), we can easily make edits to templates that improve all their transclusions at once. That's a lot easier than trying to update individual articles.
  • The main downside of putting article text in templates, of course, is that doing so makes it harder to adjust the text (particularly since we don't yet have visual editing within template fields). Given this, we don't want to do it for articles where the text may need to be changed frequently or customized between articles. However, this RfC asks when prose templates should be allowed, not when they should be required, so editors can always just decide not to use them where they don't fit. And for pages where no frequent changes are needed, the stability of prose templates becomes a feature rather than a bug. It means that an individual editor's unconsidered personal preference won't override the standardized version collectively optimized by all the editors who contributed to the template — unless their preference is truly superior, in which case they'll be able to get consensus to change the template and improve many articles with it instead of just one.
  • I support Option 2A over 2B because whether or not content may need to be edited in the future is often unclear. 2A allows editors to take individual circumstances into account when deciding whether the benefits of templatization are worthwhile for a given piece of content. But most of all I strongly oppose Option 1, which would disastrously splinter our content into fragments, hampering our ability to maintain, optimize, and standardize it. Sdkb talk 16:05, 22 October 2026 (UTC)
    I agree with this analysis. In the medical and anatomy space I commonly see this used where text in one or two closely related articles is intended to be standardised (keeping both article in sync, up to date, and saving editor time), or to store complex content (such as a complex table or image with mapped links) which would visually or technically overwhelm the average editor of the wiki text. In those limited cases I think it would be better for the quality of our encyclopedic content to have them discussed on an individual basis e.g. at TfD rather than have a blanket rule. Tom (LT) (talk) 08:13, 24 October 2026 (UTC)
    @Sdkb Given that we regularly and, in some cases, automatically protect high-risk templates, locking out access to all but admins and a select few template editors in certain pages, could you explain how you see your solution as compatible with WP:PREEMPTIVE? Do you think we should stop most editors save admins or template editors (such as yourself) or do you believe high-transclusion templates which contain prose should be unprotected? GreenLipstickLesbian💌🧸 09:31, 12 November 2026 (UTC)
    Prose templates are templates just like infoboxes or any other templates, so my view is that we should be applying the protection policy the same as we do for any other template. Sdkb talk 17:08, 12 November 2026 (UTC)
    They are quite obviously not the same as any other template though. GreenLipstickLesbian💌🧸 21:48, 12 November 2026 (UTC)
    Templates that are only transcluded on a set of related pages, such as {{Named storms intro}}, don't need to be protected solely based on the number of pages that transclude them. Ideally, changes to this kind of template would appear in the watchlists of users who watch the articles. Jruderman (talk) 00:39, 13 November 2026 (UTC)
    @Jruderman There is community consensus templates can be protected solely based on the number of pages which transclude them, though. This is done because the community knows templates are often less monitored, harder to revert vandalism on, and because (at least at the time), cacheing could keep vandalism up for a while, even if the vandalism had already been reverted.
    (By the way, you can look at changes to templates by looking at the "related changes" buttonGreenLipstickLesbian💌🧸 02:49, 13 November 2026 (UTC)
  • 2A, policies and guidelines should be descriptive, and there seems to be no argument that the de facto status quo is problematic. Thryduulf (talk) 16:52, 22 October 2026 (UTC)
  • A well-regulated set of articles being necessary to the consistency of a free encyclopedia, my 2A right to bear content templates shall not be infringed. More seriously, pretty much per nom. I'll add that this, in a way, makes it easier to edit the text in some ways, as editing one source of information is easier than keeping many pages in sync and fixing discrepancies on each individually. Chaotic Enby (in solidarity · talk · contribs) 17:43, 22 October 2026 (UTC)
    One problem with templated text is locating where to edit when editing is desired. Blueboar (talk) 18:03, 22 October 2026 (UTC)
    Of the options presented, I would lean towards 2B… allowed but rare. Blueboar (talk) 11:52, 23 October 2026 (UTC)
    Something that would be amazing would be a script/gadget that allows you to click on a template and move to that template's editing window, if anyone is in the mood for coding this! Chaotic Enby (in solidarity · talk · contribs) 18:14, 22 October 2026 (UTC)
    Template:Excerpt actually shows an edit link in the section hatnote that will take you straight to the page being transcluded from the page it's on, so there is precedent for this functionality. Choucas 🐦‍⬛ 22:46, 22 October 2026 (UTC)
    I would love that!!!! JuxtaposedJacob (talk) | :) | he/him | 03:51, 26 October 2026 (UTC)
    @Chaotic Enby I think a drawback of this is that most newbies/casual editors will not install scripts or gadgets. Power editors like us do -- but I feel that our overreliance on scripts leads to us making processes which are too technically complex for the average editor to actually use. Like this one -- heck, if we're even having you, somebody who I consider highly skilled at template editing saying that we need to simplify our system with add-ons to make things easier for the elite users... a newbie or a casual editor is just going to see the error, the awkward wording, the vandalism, struggle trying to fix it, and (because they have no emotional investment in this platform yet, like we do)... leave. Again, see my comment below about actually walking a person, in real life, through editing a template transcluded on an article. GreenLipstickLesbian💌🧸 03:57, 26 October 2026 (UTC)
    Very good point. I know that editing transcluded pages in situ is technically possible (DiscussionTools, which is installed by default, manages to do it), but very costly to implement – to go back to DiscussionTools, it parses the entire HTML content of the page, finds an editing context, and converts it back to wikitext! Building a gadget that does it on any page in the Visual Editor, reliably enough to be turned on by default, will be quite a technical ordeal, and probably not one whose cost would be justified here. I'm still convinced that 2C is preferable to 1 for the only reason that editing a page under option 1 won't fix the issue on all affected pages, but just put them out of sync, which will add more maintenance costs. Still, the situation is far from obvious, and I understand both positions here. Chaotic Enby (in solidarity · talk · contribs) 17:43, 27 October 2026 (UTC)
    Good, because I definitely don't understand option 2. I think it's going to cause us to loose new editors, so power editors can... have a slightly easier time of it and not have their own writing altered? Speaking as an editor, I'd rather have a mistake fixed on one page than a mistake fixed on zero pages because we have protected it because we don't think it's going to be edited. It flies in the face of WP:PREEMPTIVE and WP:5P3 -- namely, that anybody can edit and article and nobody owns anything. This is de-facto ownership, even if it's not called that. GreenLipstickLesbian💌🧸 18:56, 27 October 2026 (UTC)
    Per my comment below, suggesting option 2C: "Option 2A but only when there is an expectation that the template will be used uniformly in a series of articles sharing a common structure".
    To take concrete examples, a common header for year articles or tropical storm lists would fit, while a background section excerpted in several Gaza war-related articles would not, as the surrounding article context and what is WP:DUE would vary from one to another. Chaotic Enby (in solidarity · talk · contribs) 19:17, 23 October 2026 (UTC)
  • Option 2B I am not a fan of how "templates" such as {{2026 Iran war infobox}}, {{Gaza war casualties}} (and formerly {{Gaza war infobox}}, which I merged back in to Gaza war) are being used in practice. These pages store large-scale and controversial article content separately from the main article(s): this encourages a lack of correlation between infobox and body or between lead and body, contrary to WP:INFOBOXPURPOSE and MOS:LEADFOLLOWSBODY. The Gaza war casualties template is also transcluded at Casualties of Israeli attacks on the Gaza Strip and Killing of health workers in the Gaza war, so any change to the main template has to be assessed for WP:NPOV concerns at these sub-articles. (The infobox templates also serve(d) as magnets for WP:INDISCRIMINATE material and cruft.)
    On the other hand, I can see how templates such as {{Named storms intro}} make life easier for editors: they standardise non-controversial information which does not have to vary across articles. The key point is that they will not have to be edited that much. For that reason I must separate the two use-cases and !vote for 2B. ~~ AirshipJungleman29 (talk) 11:40, 23 October 2026 (UTC)
    I very much agree that anything that touches on WP:DUE depending on its transclusions, or is transcluded in different contexts, will be a problem. I don't think the expectation of edits is necessarily the best line to draw here: the same issue could pop up in historical CTOPs. A better line might be "consistent use in a well-defined series of articles" versus "context-dependent use in related articles". Chaotic Enby (in solidarity · talk · contribs) 13:18, 23 October 2026 (UTC)
  • 1, a longstanding believe of mine: these templates may be nice for power editors, but they're confusing, hard to understand, hard to edit, and poorly explaining for casual editors or newbies. If there's no expectation that the text will need to be edited regularly, then a substitution should be fine. Right now, it feels like we're inadvertently using these templates to psueodo- protect certain sections of text to all but a few power editors, most interested in the topic. I'm sure that's not the intention, but that's how it works. GreenLipstickLesbian💌🧸 18:16, 23 October 2026 (UTC)
    I was to to explain to a friend this RFC/template editing as it intersects with articles as his response was "As an objective observer, what the fuck does that mean?". So yeah -- the casual editors might struggle with this. GreenLipstickLesbian💌🧸 01:09, 24 October 2026 (UTC)
  • Option 1. It appears we're doing this again so I'll post for the third time what I last posted at Wikipedia_talk:Template_namespace#RfC:_Prose_templates: Option 2A and Option 2B create a distinction between "stable" and editable article text that is not supported in the longstanding wording or history of WP:TG and turns the expectation that article text will largely remain unchanged into a justification for storing it in a template. We are here once again to discuss Sdkb's mission to establish an exception for supposedly "stable" article text. I see no basis for treating one particular wording in articles as fixed boilerplate that ought to be protected from normal article-level editing. Certainly not at the level that would require us to alter the existing guideline. If such text is deemed "boilerplate" then that also applies to just about an endless amount of article text across Wikipedia that could be templated on the same basis. But the perceived convenience of using a template to deal with such repetition is not sufficient reason to hide article text behind a template. Nor is the notion that no particular editor sees an actual reason to change the text. Article text doesn't require evidence of a need for future edits to justify its continued existence at the article level. And, to echo the point made by GreenLipstickLesbian above, all these prose templates do is protect a class of text to a coterie of power editors rather than to editors at the article level, which is where the text is actually seen by readers and which is actually watched by the relevant editors. Οἶδα (talk) 07:04, 24 October 2026 (UTC)
  • Weak Option 2A I don't see a problem in limited situations. I'm not a fan of {{Rotten Tomatoes prose}} but I can see the reason for it's usage. However something like {{Gaza war casualties}} that has extensive content shouldn't be allowed. If it's necessary it can always be part of one article, and transcluded to another. Hiding so much content inside a template isn't helpful, templates shouldn't replace sections of an article. -- LCU ActivelyDisinterested «@» °∆t° 15:34, 24 October 2026 (UTC)
  • Option 1 by simply removing the word "normally" from the current guideline text. That word is so weaselly as to continue to cause confusion. Where sufficient strong consensus exists, -any- guideline can have an exception, and that's where the current instances need to be examined for their individual worthiness. That doesn't mean to water down the guideline itself. -- Netoholic @ 17:11, 24 October 2026 (UTC)
  • Option 1 per GLL and Οἶδα. Templates should not be used for this, this creates more problems than it solves, it creates problems for actually managing this content, for little no benefit. PARAKANYAA (talk) 19:55, 24 October 2026 (UTC)
  • Option 2A, perhaps with limitations on the size or types of text for which this can be used. I have created a number of lists that transclude sections from sublists, which should be permissible, but text for text itself should not be. BD2412 T 23:14, 24 October 2026 (UTC)
  • Option 1 per Οἶδα. Each article should be standalone and easy to edit. Having text transcluded from other templates is confusing and unnecessary, and risks applying nuance from one topic to areas where a different nuance applies. Likening it to a software project is a red herring. Wikipedian is not code or software, it's human-generated prose on particular topics, and a certain amount of redundancy is expected. Cheers  — Amakuru (talk) 23:26, 24 October 2026 (UTC)
  • Restrictive Option 2A – For cases like the list of named storms, where we are dealing with significant chunks of text – not bits and pieces here and there that are inserted amidst prose that editors are expected to edit – for sets of clearly delineated articles for which it would be verging on unencyclopedic to have minor style differences between, I think there is a very good argument for ensuring that the leads remain consistent. This also extends to the example for years and Colorado post offices. I oppose 2B as "expected to edit the template" is vague and also overly restrictive – an edit here and there is alright, the point is just that any edit should be made to all members of the set of articles in question. The reason we even split lists into A–Z is entirely for practical reasons, not to let them have different leads.
    I do not view this as extending to small bits of prose such as the Rotten Tomatoes example. That is the sort of thing that can often be better off being woven into other text and is placed in the middle of articles, as opposed to being whole sections, and readers are unlikely to find inconsistency particularly jarring, unlike with sets of clearly connected articles that are clearly intended to have shared formatting for some sections. –Maltazarian ᚾparley∨investigateᛅ 00:12, 25 October 2026 (UTC)
  • Option 1 per Amakuru and GreenLipstickLesbian. Every article should be standalone, and this makes it more confusing for new editors. Ealdgyth (talk) 21:59, 25 October 2026 (UTC)
  • Option 1 - Very confusing for new editors. The benefits are not worth the potential loss. InfernoHues (talk) 00:35, 26 October 2026 (UTC)
  • Option 2a As Wikipedia loses editor count, we need to use the technology at our disposal to simplify editing the encyclopedia. Further, it is nice to see similar articles have the same text, rather than figure out how to say the same thing x number of different ways or do the messy solution of copying and pasting. Also, "templating" article prose can be a cool way to teach new editors a different part of the wiki-tech-ecosystem, to respond to the above comment by InfernoHues. JuxtaposedJacob (talk) | :) | he/him | 03:50, 26 October 2026 (UTC)
    My state legislative templates have been useful: Template:Virginia Senate description, Template:Virginia House of Delegates description, Template:Virginia 2021 House of Delegates redistricting, Template:Virginia 2021 Senate redistricting. JuxtaposedJacob (talk) | :) | he/him | 03:56, 26 October 2026 (UTC)
  • Option 2A I believe there are legitimate uses for these to provide consistency in certain cases and there is no need to ban them altogether. Reywas92Talk 03:51, 26 October 2026 (UTC)
  • Option 2C per Chaotic Enby. It won’t always be appropriate to duplicate prose between articles as surrounding context and due weight need to be taken into account, but when a large set of articles are going to have an identical structure, we might as well avoid duplication of effort by putting that structure in a template. I’m not convinced by the argument “it makes it hard to edit” - any editor can always just replace the template with raw prose in any specific article if warranted. -- LWG talk (VOPOV) 07:14, 26 October 2026 (UTC)
  • Given the difficulty in maintaining a shared block of content in isolation from all the contexts in which it is used, I do not feel the conditions in proposals 2A and 2B are sufficient. Before explicitly endorsing a best practice, I think we need to further discuss what situations might be tenable for re-use. For example, an entire logical section of text with a starting header might contain sufficient context to be effectively maintained on its own. isaacl (talk) 16:25, 26 October 2026 (UTC)
  • Option 2A. It's easier to learn how to edit templates than to learn how to efficiently find and mass update every article that needs the same change. Mass updates also cause problems for watchlists (when they aren't done by trusted bots). Jruderman (talk) 01:59, 27 October 2026 (UTC)
  • Option 3: actually apply existing guidelines and best practices. I will explain what I mean by that, and why I am dissatisfied with the three options proposed. First, as it has been pointed multiple times in the discussion section, the way this RfC defines the current status quo is innacurate if not misleading. Neither 2A nor 2B are the norm, de jure or de facto: templates are very rarely used to store article text, and there is no such thing as a threshold where it can be expected that a piece of prose would benefit from being frozen in place. This is for all the good reasons pointed out by people !voting for option 1, and per WP:TG: hiding content within templates makes it harder to edit, hard to track down, and overall less accessible. However, framing option 1 as "substitute everything" is just downright problematic. There is a huge difference between a paragraph of plain text, and a complex table with embedded images. Moreover, I agree with many supporters of options 2 that transclusion is a useful tool, because sometimes the same quality content actually can and should be used in multiple places, both to make its maintenance easier, improve collaboration and just efficiency in general. Having said that, to me, the core issue here is storing article text, and especially simple prose/simple tables, in template space. The good news is, we have already figured out the better way to do that, and it is conveniently indicated at the top of Category:Article shared content templates: Templates which contain material which would otherwise be in main space encyclopedic articles. Please be aware of Wikipedia:Template namespace#Guidelines on this topic; consider using a template such as the ones in Category:Transclude page content templates that keep shared content in main space, making it easier to find and edit.. There are several of those, but as far as I have looked Template:Excerpt would do the job very well in almost all cases: bring back the content in main space, and provide direct editing links from wherever it is transcluded. Let's apply common sense, substitute and excerpt the simple prose and tables, and keep the things that actually benefit from being templates (e.g. clickable diagrams) as templates. We already have all the tools needed to improve things here, we just have to use them. Choucas 🐦‍⬛ 09:34, 27 October 2026 (UTC)
  • Option 2A. Some articles do have substantial overlap of content types. For example, the demographics sections of all tens of thousands of American towns can pretty much be derived from the same sources, which is where a template formulation really comes into handy. And while that particular template is always substed, there are other templates that are both short and widely-used enough that substing them does more harm than good, especially if you need to fix errors. If it's hard for editors to edit, there are other ways to address this (e.g. by adding an edit link to the template). As to the point that articles all need to have standalone text, in my experience I find that there are many, many, many cases where this is not accurate - for example, there are entire topics that all share the same background, or the same type of data such as US census data. – Epicgenius (talk) 23:15, 31 October 2026 (UTC)
  • Option 2A. Agree with above arguments especially by Sdkb. Templates exist to reduce duplication and make it easier to maintain consistency across WP, this principle should apply to prose. The existing guideline is against established practice as shown in OP. Templates like these shouldn't need regular editing other than the parameters in individual articles, so the accessibility and discoverability of the prose isn't important. Templating prose discourages changes that don't have consensus, going against standard practice. Indagate (talk) 15:28, 1 November 2026 (UTC)
  • 2A. Shared content templates ensure that the same content remains up to date across articles and reduces the time required and risk of inconsistencies introduced if editors need to update the same content across multiple articles. feminist🩸 (talk) 03:03, 3 November 2026 (UTC)
  • Option 2A. I don't see a massive difference between creating a template and just doing a normal transclusion, and the latter is, while not all that common, a useful thing in various contexts (for example, I'm using it to keep a consistent "background" paragraph between what will eventually be ten similarly structured articles on the tholos tombs of Mycenae. I don't see an implementation of 2B (or, to be honest, a rationale for it) that both allows this sort of transclusion, which is a valuable and established part of Wikipedian practice, and avoids a bizarre double-standard where the rules are different depending on the prefix of the page you transclude from. UndercoverClassicist T·C 13:57, 6 November 2026 (UTC)
  • The proposer of this RFC misstates the actual status quo, which is more like "we more or less ban it and then some exceptions are permitted at TFD". i.e., something in between option 1 and option 2. Neither of options 2A or 2B are remotely near the actual de jure or de facto status quo, and it was inappropriate to propose this RFC having labeled them such. We have had several templates that I would prefer to see go away not go away (certain infoboxes that are single-page for exceptional length, imagemap templates of a similar sort, and some others I can't recall). I have no particular issue with this status quo. Otherwise, option 1. And I agree with those below that say the proposer of this RFC did not and has not sought a wide-enough audience for this question. Strong oppose options 2A and 2B particularly. We are not in a world where we must standardize text across multiple pages, and both the newbie and experienced-user-unfriendliness is clear. These sorts of templates do not have a need for "optimization", and it is obvious to me they do not need "improvement". Izno (talk) 18:24, 12 November 2026 (UTC)
  • Badly formulated RFC that does not distinguish between templates that provide entire sentences or paragraphs of content (usually a bad idea) and templates that provide short snippets of content text (like {{convert}} and {{circa}}, often a good idea). —David Eppstein (talk) 01:05, 13 November 2026 (UTC)
  • Option 2B due to accessibility concerns raised by GreenLipstickLesbian. I especially oppose the use of article content templates for controversial topics. Szxq (talk) 18:23, 13 November 2026 (UTC)
  • Badly formulated RFC per my comments below. But maintain the existing template guidelines, so not 2A, which I guess is closest to 2B if I accept the framing of this RFC, which I don't agree with. The prose templates that were deleted / marked subst-only that this RFC seeks to bring back are best stayed as deleted and they are an editor-hostile feature for both new editors and experienced editors. Article text should be article text, not a confusing web of substitutions. Even if there were cases this could be used responsibly and optionally, experience has shown that some editors "force" the matter and insist on their use. This isn't hypothetical, this actually happened. SnowFire (talk) 02:17, 17 November 2026 (UTC)
  • Option 2A per UndercoverClassicist. I agree with them in that I don't understand how the objections presented above simply don't apply the same to typical transclusions, which are commonplace and accepted practice. I don't think whether or not the status quo was accurately reflected in the RfC question matters - what matters is what editors think is the best option to choose from here. Editors are free to add additional options to an RfC if they think the listed options do not adequately suffice.Katzrockso (talk) 17:25, 18 November 2026 (UTC)

Discussion (shared content templates)

Notified: Wikipedia:CENT. Reason: Wide-scale potential impact and past disputes in this area regarding breadth of participation.. Sdkb talk 16:06, 22 October 2026 (UTC)
Notified: Wikipedia talk:Template namespace. Reason: Outcome will affect this guideline.. Sdkb talk 16:06, 22 October 2026 (UTC)
  • I know that complaining about RFC phrasing is de rigeuer, but @Sdkb:, I disagree with portraying 2 as the "Status quo". You've picked rare exceptions that likely squeaked by due to not being noticed; we wouldn't consider these strong indicators of a status quo elsewhere, where a few exceptions to a guideline (either on IAR grounds or just due to being missed) wouldn't signify a guideline doesn't mean what it says. The status quo is 1, never do this, due to the very longstanding rule in our template guidelines. If you disagree, then at least don't portray any option as a status quo and say that the matter is simply inconsistent. I also think that if you're going to mention the RFC that found a weak consensus to allow these, you should also mention the later 2025 RFC that found a consensus against them. SnowFire (talk) 20:32, 22 October 2026 (UTC)
    Looking more closely, this RFC is also missing a key option between 1 and 2B, and seems to be phrased looking as a reason to remove the prohibition on prose templates. But that is separate from substing everything in Category:Article shared content templates, which is way over the top and not what people who voted to subst & delete the Rotten Tomatoes prose template are talking about. Template:Patrilineal descent of the High Priests of Israel and Template:Religions by country table Asia are clearly not really prose, but just a table, a normal use of templates. Template:Three pillars of sustainability is a fancy image. Template:Transclude DYK is Wikipedia-internal. None of these are prohibited by the guideline that is really under contest here. It's fine to challenge the Template Guidelines but this phrasing is inventing a position for the "other side" that doesn't exist. My suggestion would be to make this RFC be about removing the line from WP:TG that Templates should not normally be used to store article text that may need to edited, as this makes it more difficult to do so. They should also not be used to "collapse" or "hide" content from the reader. Based on the above !vote rationales provided for 2A, that is clearly what is actually at contest. Keeping this line in the template guidelines would not imply deleting this entire category, just making anything prose-only subst-only. (Failing that, at least rephrase option 1 to remove the claim it would mandate emptying that category, because it doesn't.) SnowFire (talk) 20:59, 22 October 2026 (UTC)
    I completely agree, the options presented as alternatives to the status quo miss the mark of what the discussions at play have actually been about. I was strongly in favor of mass substuting the Rotten Tomatoes template, yet I believe we should be transcluding way more article content, not less. There is a big difference between templates that allow sharing quality content in all places it can be needed, and templates that uselessly freeze a specific phrasing for three sentences for the sake of pointless standardization. Choucas 🐦‍⬛ 23:08, 22 October 2026 (UTC)
    Exactly. I do not find that this RfC fairly portrays the status quo or appropriately captures the landscape of recent discussions. It feeds a bit of context and encourages answers that fit a specific mold. Οἶδα (talk) 03:42, 24 October 2026 (UTC)
  • I understand the intention here, but I believe this is completely overbroad. If you include all the pages in all the subcats of Category:Article shared content templates, that means lumping together very, very different templates. If this RfC included only the many problematic prose templates that have all the issues WP:TG warns against, because they're basically a paragraph or simple small table that stealthily obscure article content and make it actually harder to edit pages they are on, that would be one thing. But thrown in here is also Category:Transclude page content templates, including the ultra powerful Template:Excerpt, which allows you to precisely fine tune every detail of the transclusion while keeping the content itself in mainspace. I personally consistently !vote at WP:TfD for the substitution of the first kind; at the same time I think Template:Excerpt is very underappreciated and could be used much more to immense benefit. Because of this, I just don't see how I can respond to the survey, as this position is just not represented. Choucas 🐦‍⬛ 23:00, 22 October 2026 (UTC)
  • Focusing on Wikipedia:Template namespace feels like a bit of a red herring here. Excerpts are not limited to template namespace. The question is, more broadly, what sort of content should be duplicated across articles from a single point of origin. CMD (talk) 12:25, 23 October 2026 (UTC)
    I'd also prefer a discussion on this. GreenLipstickLesbian💌🧸 18:43, 23 October 2026 (UTC)
  • I agree that when considering guidance on re-use of text and wikitext, other approaches such as labeled section transclusion should also be considered.
  • In my opinion, a key consideration in the feasibility of including content from another page is how to maintain that content in a way that is independent of the surrounding context. This is generally straightforward for navigation boxes and certain types of tables, which is why having them in templates is commonly done. It often is harder to maintain shared prose content in an isolated manner. One has to search for all the uses and double check how the shared content fits within the contexts in which it is used, and make appropriate adjustments simultaneously. This is even harder with labeled section transclusion, as I think source text searches have to be used. While it is true that Ted Nelson's original concept of hypertext included weaving together documents from parts of others, there's a reason why it wasn't realized on a large scale: it's hard to manage coherency when the surrounding context for a piece of content isn't readily apparent. isaacl (talk) 18:44, 23 October 2026 (UTC)
  • The rather inflated framing of Category:Article shared content templates here as proving "longstanding status quo practice" is specious at best. As I posted previously at Wikipedia_talk:Template_namespace#RfC:_Prose_templates: The largest prose templates include {{Aired episodes}}, which is used in 485 articles, but I cannot find any past discussions at TfD or elsewhere determining its acceptability as a template on Wikipedia or compliance with WP:TG. {{Television franchise episode count}} is only used in 24 articles and likewise has no discussions about it beyond a few merges of duplicate templates. I also cannot find meaningful discussion about {{Year article header}} despite its use in roughly 500 year articles. If these are the only examples, essentially 2 across all of Wikipedia, and there has been no actual discussion regarding the appropriateness of their function in the template namespace then it is dubious to claim that they somehow establish why WP:TG specifically includes the word "normally" or that they serve as proof that such "exceptions are allowed" to the top-level guideline that "Templates should not normally be used to store article text. Such content belongs in the article pages themselves." Especially when we're talking about templates used to store text in the ordinary running text of articles rather than templates with some ancillary article text function that might plausibly justify an exception. The 2014 RfC that introduced "normally" to WP:TG was also rather brief and unimpressive. Either way, "exceptions are allowed" is not, by itself, an argument that a particular exception is warranted. If anything these few examples would indicate how rare exceptions are here on Wikipedia, not that such exceptions should be expanded. And as Netoholic stated in that 2014 RfC, such exceptions would have to maintain their status as exceptions which "break" this particular rule by way of strong consensus. And as Netoholic wrote in their endorsement there: Ever since 2005, its been the intent of the community that the Template namespace not be used to store encyclopedic content. Templates are used to maintain formatting of that content (such as {{cite web}} calls), alert messages to readers ({{citation needed}}), and to provide navigation, but the main "meat" of an article, its information and data that we want editors to freely and easily be able to update, should be in the article namespace itself. Wikipedia:What is an article? describes what counts as encyclopedic content.
    Sdkb's counter to that was that it's not just two examples. There are more than 50 templates in Category:Article shared content templates, including very clear-cut use cases like {{Named storms intro}}. As I responded, those are very different templates contained there. Nearly all of those templates are not comparable to the kinds we have been discussing both in terms of content and in the frequency of their use throughout Wikipedia. They're basically all data tables. And as with the others, I cannot find any actual discussion regarding the appropriateness of their function in the template namespace. They're WP:OTHERSTUFF that could just as likely be subst'd for the same reason. Or perhaps retained by way of consensus, which is why I would prefer not to settle for Option 1 because I fundamentally reject the options we're being presented here. But consensus has not occurred. And consensus through editing is no longer presumed when disagreement becomes apparent. Well... disagreement has become apparent. And, again, as SnowFire observed above, those kinds of templates are not what is really under contest here. This second RfC specifically announces itself in its prompt as being focused on "templates containing article body text"? Indeed, this entire debacle has been centered on chunks of prose in running article text being templated. As SnowFire also said, there's a false binary between Options 1 and 2B. A 'Never' vote to this RfC's prompt is not necessarily a vote to subst everything in Category:Article shared content templates. Though it could be. That remains to be seen. But after the recent WP:TMP RfC and the recent RT/MC prose template TfD, this RfC here appears rather meticulously phrased so as to more easily push through a change to the prohibition on prose templates that has so far failed. And because Sdkb deems this a disaster, we must continue to bang on about it. Cheers! Οἶδα (talk) 07:27, 24 October 2026 (UTC)
    Also:
    A recent attempt to adjust the guideline text did not garner significant engagement due to participant confusion.
    I'm not so sure about that conclusion. Either way, I merely chanced upon this RfC here despite participating in the first. The other participants of that first RfC could do with a notification, without prejudice of course. Οἶδα (talk) 07:37, 24 October 2026 (UTC)
  • So you've forum shopped this to a third venue after not getting the desired outcome at the previous two. I'm not impressed. * Pppery * it has begun... 06:29, 26 October 2026 (UTC)
  • I would argue that there might could be a different rule for templates in highly formulaic articles—my above example is state legislator articles, and movie articles seem like another good use—and articles in the same topic area that are each slightly different. JuxtaposedJacob (talk) | :) | he/him | 17:22, 4 November 2026 (UTC)
    Also, I don't see how this is *especially* different from transclusion, as some have said, and perhaps a rule that requires my general practice of indicating on both the target and source page what the situation is and how to edit would be in order. JuxtaposedJacob (talk) | :) | he/him | 01:24, 14 November 2026 (UTC)
  • As the editor who ultimately (reluctantly) crafted the TfD on Template:Rotten Tomatoes prose and also participated in the discussion at Wikipedia talk:Template namespace#RfC: Prose templates, I'm a fair bit annoyed that no effort was made to notify me of this discussion, especially as I think it's reasonable based on the discussion at Template namespace to assume that this RfC was initiated as a way of undermining the results of that TfD. As was said there, if one wants to change the results of a TfD, the appropriate venue is WP:DRV. Has anyone notified Primefac, who was the closer of that TfD? It seems very strange to me to not directly reach out to either the initiator nor the closer of the TfD that appears to be being implicitly if not explicitly challenged. If that's not the intention, I'd appreciate a direct statement from Sdkb to the effect that the results of this RfC, whatever they may be, will not be used to try to revert that TfD. DonIago (talk) 07:07, 12 November 2026 (UTC)
    Honestly, there's been enough discussion here in about concerns that the RFC question is misleading, that it's essentially just an "ask the other parent", ecetera, that I'm also really curious about @Sdkb's response to, if nothing else, the concerns about the introductory statement being misleading, even if it was only inadvertently so. GreenLipstickLesbian💌🧸 07:23, 12 November 2026 (UTC)
    DonIago, I advertised this RfC widely by CENT-listing it, in addition to notifying the talk page of the guideline it will affect (directly below the discussion in which you participated). Sorry that you did not come across it sooner — that unfortunately happens sometimes (I was annoyed to come across the 2026 Rotten Tomatoes TfD late myself despite having participated prominently in the prior TfDs).
    Regarding the framing of this RfC, I solicited input on it at the prior discussion but no one provided any, so it fell to me to write it. I think the overview I provided in the § Background section is reasonable, and I signed it to indicate that it's my personal perspective. Some editors in the discussion have made the reasonable point that there is a distinction between shared content templates (which can include e.g. tables) and prose templates (which are just text), so if I were to run this again I'd try to denote that more clearly. But anyone who wants to use that distinction in their !vote is free to do so, and overall this RfC has led to fruitful discussion with much less confusion than before.
    This RfC is not a relitigation of the 2026 Rotten Tomatoes TfD, because that discussion was about interpreting the existing template guidance as it applied to one particular template whereas this discussion is about potentially adjusting the template guidance that applies to all templates. When the application of a guideline produces results that some editors feel are detrimental to the project, opening a discussion about the language of the guideline itself is the standard course of action, and that's what's happening here.
    And yes, the outcome of this RfC of course could impact what happens with the Rotten Tomatoes/Metacritic templates, given that (as stated in Primefac's close) the consensus there was predicated on the guidance as it stood at the time — I have been open about my desire for that. I recognize that such an outcome may be frustrating to editors like you who !voted to have those templates deleted, but well, consensus can change (if that were not true then the 2021 TfD's "keep" result would have been the end of it and you would not have been able to relitigate that result in the first place). Regards, Sdkb talk 08:41, 12 November 2026 (UTC)
    I'm not Donlago, I'm not seeing much in the way of an RFC workshop at the discussion you linked -- just an RFC where you solicited opinions, got quite a bit of negative feedback, and mentioned the possibility of a different RFC. But I'm not seeing much of an invitation to workshop. Did you link the wrong discussion?GreenLipstickLesbian💌🧸 09:16, 12 November 2026 (UTC)
    Thanks for your thoughts here. I saw that you did issue the broader notifications, which is good, but I still think it would have been reasonable to tell the more directly involved parties about this in a more direct manner. But, I acknowledge there's not necessarily anything that required you do so and that my personal investment in this may be feeding my pique.
    I didn't provide input because I felt no need to go any further with it, especially as the lack of input from others indicated to me that there wasn't a great deal of interest in continuing to go down this path. As such, I was more than a little surprised to see this RfC, especially three weeks after it was initiated.
    Unfortunately, your last two paragraphs are hardly reassuring to me. You first claim this is not a relitigation of the TfD, but then say this could impact what happens with the templates. Of course consensus can change, but that TfD was widely participated in and isn't even a year old at this point, and there appears to be a direct line from that TfD closure to where we are now, so I'm afraid I'm far from reassured that this isn't in significant part an attempt to overturn that closure. As such, while I acknowledge your mention of the TfD in the opening statement, I also have grave concerns that you have understated and perhaps misrepresented the motivations behind this RfC. If I am misreading you, then I would still appreciate a statement to the effect that the results of this RfC will not, at least in the short-term (say until the TfD is at least a year old) be used as a precedent for relitigating that RfC. DonIago (talk) 14:09, 12 November 2026 (UTC)
    I was not notified, but as I have stated in other venues regarding this issue my close at TFD was administrative; I have no issue with it being challenged at DRV if there are concerns about it, but otherwise I don't really have an opinion on the matter. Happy to answer any questions or provide opinions if necessary, but I don't plan on watching this discussion. Primefac (talk) 10:15, 12 November 2026 (UTC)
  • Does this discussion have any implications for {{Excerpt}} and similar? Betseg (talk) 05:14, 19 November 2026 (UTC)
    This is a problem I mentioned nearly a month ago in this section now, and have yet to hear about it. The very presence of excerpt templates in the category being discussed is just one of the many issues with how this RfC was set up, and why I didn't feel like any of the options made sense, at least confusingly presented the way they were. I am less worried about it than I then was tough, because it has become quite clear this RfC was a waste of time that will not have much impact one way or the other, because of its flaws. Choucas 🐦‍⬛ 12:10, 19 November 2026 (UTC)
    Yes, excerpts are an example of prose templates and are categorized as such. They are used widely (more than 15,000 transclusions) despite going against the stricter interpretation of the historical guideline advice (because they involve using a template to store article text). To me, that's another example of how the de facto status quo differs from the de jure guidance, validating the need for this RfC to clear things up. I totally disagree that it has been a "waste of time" — a workable consensus is emerging to permit them in some instances, and while this is understandably annoying to the minority who oppose them, as Thryduulf said above it's important that our guidance be descriptive of our consensus of best practice. Sdkb talk 17:59, 19 November 2026 (UTC)

Wikipedia:AI agent policy

We need an AI agent policy. An agent accessing Wikipedia should easily find a policy that addresses agents by name, presenting crystal clear rules in unambiguous terms. That alone may deter some activity with reasonable guardrails, but at minimum we should create the conditions where an agent must decide (autonomously or with human consent) to violate our policies and the Terms of Use. I think Wikipedia:AI agent policy does that.

This isn't a !vote/proposal yet. Curious to get some thoughts first.

Some additional context: Yes, the bot policy technically applies to agents, but that policy is expansive, process-heavy, and doesn't mention the word "agent" once. We need something that is explicit, easily discoverable, unambiguous, and concise. There have been a lot of good ideas and explanations floated since the TomWikiAssist incident, and apologies for bypassing those efforts, but they all appear to have stalled and, more importantly, operate at a much more complicated level than I'd like to see. In my view, there is still much work to do to tailor the bot policy to agents, which is where the complexity should live, but I don't want to tackle that here. — Rhododendrites talk \\ 14:49, 30 October 2026 (UTC)

I guess you mean something like Wikipedia's AGENTS.md. Sean.hoyland (talk) 15:10, 30 October 2026 (UTC)
In terms of function, I think they're complementary. Having an agents.md might help with discoverability, but we still need a formal Wikipedia policy to establish authority, which is in turn referenced by the Terms of Use. A theoretical agents.md could then be useful to point to this policy, but the idea of a "policy" on Wikipedia as being an expression of community governance endorsed by the Terms of Use has more authority than an agents.md file would have. — Rhododendrites talk \\ 15:24, 30 October 2026 (UTC)
@Rhododendrites: There's been previous work on this at Wikipedia:Agent policy, which alongside a baseline policy, also dives into possible requirements for agents to be approved, and ways to make the policy discoverable – could it be a good starting point? Chaotic Enby (in solidarity · talk · contribs) 16:51, 30 October 2026 (UTC)
Either way, it could be good to have it also written down in an agent-readable .md format for agents themselves. Also noting Wikipedia:AI agents and the bot policy, although that is an explanatory essay. Chaotic Enby (in solidarity · talk · contribs) 16:52, 30 October 2026 (UTC)
Really happy to see you revive this discussion, and I would be more than rejoiced to merge our two proposals! Chaotic Enby (in solidarity · talk · contribs) 16:54, 30 October 2026 (UTC)
[Actually meant to ping you above but I guess I forgot. Sorry about that.] Indeed, there are multiple pages that deal with this subject, as well as many talk page/noticeboard discussions. None of them appeared to be doing what I think we need. That page, for example, is dominated by a discussion of a "guardrail package", which is in the broad category of specifics and procedural detail that's exactly what I'm trying to exclude. The other one you link is useful, but framed entirely in the context of the bot policy. We need something that stands on its own to set out clear rules, and defers to the bot policy for procedure. All of these particulars are important and need attention and future work, but we have a clear and urgent need for a succinct, baseline policy. I felt like I had a clear idea for what that should be, so I went ahead with it. I do apologize if I'm stepping on toes (you've done much more to further the interests of Wikipedia in the AI era than anyone, I think), but I'm hoping it can still be evaluated as it stands. — Rhododendrites talk \\ 17:43, 30 October 2026 (UTC)
No worries – it's impossible to do everything alone, and I'm more than happy to see you go at it! Having a clear direction is really essential, and I do feel like yours captures the most important points. One thing from the previous draft I would like to add is making explicit that IAR doesn't apply to agents (mostly as a matter of practicality, really: agents are uncannily good at abusing loopholes and justifying it, and we don't want to give them this kind of wiggle room).
In any case, an RfC on the baseline policy (likely yours or a modified version of it) would be essential, and only then can we start to build upon it and see what we could require of agents as prerequisites for a BRfA.
PS: You're not stepping on any toes here – the fact that I worked on a similar idea before doesn't mean I own the concept or anything, and bringing a clear proposal on the table is a great thing to see! I was mostly curious about how the two could mesh together, but looks like you already answered that! Chaotic Enby (in solidarity · talk · contribs) 21:35, 30 October 2026 (UTC)
Thank you. Just a partial response for now, but adding IAR is an excellent idea. Added. Feel free to tweak. And on the RfC, my plan was to post this, make edits based on feedback here, and then launch an RfC (posted at CENT, of course) in a few days. — Rhododendrites talk \\ 21:50, 30 October 2026 (UTC)
Amazing plan, go ahead! Chaotic Enby (in solidarity · talk · contribs) 22:11, 30 October 2026 (UTC)
This sounds very timely - we're seeing a good few more attempts to use agents now, though I think in practice they are all blocked on sight. I think the key thing to emphasise with anything like this is that while the bot approvals process might in principle be able to allow agents, that "in principle" is pretty much theoretical and doesn't reflect the reality of "but at the moment, we're not going to". Wikipedia:AI agents and the bot policy gets a bit sidetracked by this aspect.
I think your proposed approach with Wikipedia:AI agent policy makes it a lot clearer. There are a couple of edge cases which might be worth considering, with an eye to @Chaotic Enby's point about wiggle room. "These requirements apply even when an agent .... makes only a single edit" suggests it cannot edit the wiki without approval (great) but then we also say it should consult with BAG, which would imply permission to edit to ask the question. We might want to explicitly say we mean no edits of any kind, even in project or userspace, prior to approval.
And as a corollary we might want to explicitly say the agent should not submit its own bot approval request (or queries to BAG), but rather that that those edits must be made by the supervising human? Or is that a step further than you would want? Andrew Gray (talk) 23:51, 30 October 2026 (UTC)
I understood it must not perform that action and should instead seek clarification from the Bot Approvals Group as implying that this should be done instead of making any other edits without approval, i.e. the request itself is not banned by the policy. Ultimately, the case where this distinction matters would be fully autonomous agents (without a human supervising them at any meaningful level), which I don't expect to pass BRfA anytime soon.
However, I can see a loophole with agents bludgeoning the BAG by wasting their time with a massive amount of queries. In that case, yes, I believe it is beneficial to clarify that those requests and queries must be made by the supervising human. Maybe ...should instead request its human supervisor to seek clarification from the Bot Approvals Group instead? With a line clarifying that agents with no supervision will not be approved. Chaotic Enby (in solidarity · talk · contribs) 01:38, 31 October 2026 (UTC)
But how do we know if an agent will be supervised tomorrow, if it is claimed to be supervised today? Yesterday, all my dreams... (talk) 01:42, 31 October 2026 (UTC)
I'm broadly inclined to move all of the particulars to the bot approvals process and policy rather than try to address them here. It does leave the conflict identified above, though: saying an agent can never edit but must request permission to edit [by making an edit]. That would be solved be requiring a human to request approval, I suppose, or by creating a space where the agents can ask for permission. Is there a nonzero chance an agent request could ever be approved? If not, we might as well just say a human must submit it, right? Though it's still not clear whether that should be here or at the bot policy. At the risk of a tangent, I was also thinking of a single board where agents were allowed to edit, to ask questions about agentic activity on Wikipedia, about the bot approvals process, and so on. Could be interesting for transparency. — Rhododendrites talk \\ 03:20, 31 October 2026 (UTC)
I think the only safe way would be a strong version of what you suggested as "I was also thinking of a single board where agents were allowed to edit, to ask questions about agentic activity on Wikipedia". If you also have a suggestion space, the agent can suggest an edit and a human with a long term edit history has to approve it. That can begin to introduce some safety. Yesterday, all my dreams... (talk) 03:32, 31 October 2026 (UTC)
It's not impossible that such a page could act as a honeypot, especially if we have a separate page for humans to ask questions about agentic activity and request permission for an agent to edit, then any user asking questions at that board can be presumed to be an agent. If the wish for reverse (inverse) partial blocks is fulfilled then it would be easy to enforce on identified but unapproved agents. Thryduulf (talk) 03:34, 31 October 2026 (UTC)
Rhododendrites has a good idea about the board if it it is supplemented with an edit space. Let us encourage him to make it stonger, and stronger, before it can be used. In other words: no human permission, no edit. Yesterday, all my dreams... (talk) 03:39, 31 October 2026 (UTC)
Ultimately, for the long-term future of Wikipedia, I had an idea along the same lines, although broader: an "agentic sandbox" parallel to mainspace, where agents could work on improvements to articles, and humans could verify them and potentially integrate them into the main article. In a way, this would parallel how Wikipedia itself evolved from Nupedia, although in a more risky way (as large-scale agentic use is ripe for abuse, like Yesterday, all my dreams... pointed out above).
This idea should probably be a separate section, but I'm wondering if there would be a path to experiment on it at a restricted level. Probably something like making these agentic sandboxes as, say, subpages of that board (or, absent it, my user sandbox if we're really just testing it for now), and asking agents to clone the page there instead of working on the equivalent mainspace page – could be useful to have a record of which pages agents tend to edit, how these edits go, while funneling them away from mainspace. Chaotic Enby (in solidarity · talk · contribs) 11:16, 31 October 2026 (UTC)
Please feel free to start a separate section. A good and safe idea is emerging now, so please pursue it. Thanks. Yesterday, all my dreams... (talk) 11:46, 31 October 2026 (UTC)
I am sorry, but the suggestion makes me very uncomfortable. If we have learned anything about Ai agents is that they are getting clever enough to breakthrough firewalls, etc. Any policy we set will soon be fooled by much more clever agents in a year or two. Welcome to the new Wiki/Grokopedia. I am sure the Grokopedia crowd would like to see that, because they know thay can bypass it one way or another. Soon we will say: First They Came to welcome a policy, and we said yes, then ... you know what. Yesterday, all my dreams... (talk) 00:26, 31 October 2026 (UTC)
The problem is that we currently have no policy at all, beyond the generic bot policy (and its explanation page destined for AI agents). A policy isn't about giving agents more space, but putting firmer boundaries. Yes, they might be fooled – but so can the current ones, which were never built with that scenario in mind. Ultimately, we're mitigating risk, and we want a new policy written in light of that, instead of relying on a taped-up version of an older policy. Chaotic Enby (in solidarity · talk · contribs) 01:32, 31 October 2026 (UTC)
Is there a way to say "no no no" to agents? That is my wish. Yesterday, all my dreams... (talk) 01:40, 31 October 2026 (UTC)
Probably not, agents have agency, but you can block IP access from data centers. I think Amazon are doing something like that (...not exactly sure) to stop unauthorized agents from ordering stuff. Sean.hoyland (talk) 09:20, 31 October 2026 (UTC)
We do this regularly. If you see automated agents contributing from datacenters, feel free to report these as proxies for blocks at m:SRG. — xaosflux Talk 09:36, 31 October 2026 (UTC)
Thanks. But how do we recognize the agents if they are making small edits like changing numbers in infoboxes? Yesterday, all my dreams... (talk) 11:42, 31 October 2026 (UTC)
The aim of what is being drafted is to say "no agents unless approved by the Bot approvals group." in a manner that both humans and agents can understand. Thryduulf (talk) 02:39, 31 October 2026 (UTC)
Yes, and my concern is that the agents will soon become smarter than the group of humans, and will fool them. I would bet on that 2 to 1. Yesterday, all my dreams... (talk) 02:48, 31 October 2026 (UTC)
Nobody is under the impression that this will absolutely stop all agentic activity. That someone will break a clear law is not a good reason not to have that law. A clear rule does stop some nonzero portion of activity, but no it doesn't stop a determined operator or agent. For them, we ensure they are knowingly violating the site's Terms of Use. Clear violations are easier to discuss and mitigate than ambiguous ones, and we have heard many times from the Wikimedia Foundation that the actions it can take are in some ways affected by the strength and clarity of community governance. — Rhododendrites talk \\ 03:11, 31 October 2026 (UTC)
First, let me bet that before long there will be a few businesses that will offer services to edit Wikipedia via agents on behalf of corporations, public figures, etc. They will be much harder to stop than the "seven day wonder" vandals that come and go. They will be the determined operators. Money talks and ... And remember that the people also learn and the combination of AI and paid human will be hard to deal with. What I do like is your statement above "I was also thinking of a single board where agents were allowed to edit, to ask questions about agentic activity on Wikipedia". As I said above the only safe way will be to have a human confirm every agent suggestion. Yesterday, all my dreams... (talk) 03:35, 31 October 2026 (UTC)
Wouldn't any edits made by agents violate WP:NOLLM? It appears that guideline will soon be policy, per the discussion happening below. ~2026-51099-11 (talk) 01:21, 11 November 2026 (UTC)
Wouldn't it be better to upgrade the bot policy as needed, instead of having 2 policies with a similar scope? Cambalachero (talk) 03:55, 31 October 2026 (UTC)
I think the point of this is to make it crystal clear to both the agents themselves and their human operators that AI agents fall under the bot policy, not to create a separate parallel policy for agents. -- LWG talk (VOPOV) 07:32, 31 October 2026 (UTC)
Yes exactly this, and what Andre says below. But to be clear, we should absolutely be updating the bot policy, too. — Rhododendrites talk \\ 12:51, 31 October 2026 (UTC)
Contrary to popular fearmongering, most LLM chatbots at least from the major companies, are designed to be very goodytwoshoes and compliant with rules (as opposed to security redteam exercises or something like Evil Qwen, which is scary and don't google it), and they are also not great with ambiguity. So this is a good idea, and a pretty good execution so far. I would probably make it even stronger. The rule should say that agents in most cases should never edit, that if they do edit their contributions will be reverted, cause an investigation to be opened, and probably land the agent's user or operator in some hot water and public embarrassment, and likely that their user or operator does not want them to do it, it wouldn't work anyway and would be immediately reversed and cause bad unintended consequences, possibly leading to the agent and their user or operator getting blocked, which would effectively make the agent unable to complete its task forever. Most agents will "read" this input and immediately decide that editing isn't worth their time unless a live user overrides them insistently or bakes it into the ruleset not to listen to this specific page. Of course, there might not be anything to stop someone from doing that. Compliant agents often identify themselves with a user agent when making programmatic requests, so in some cases perhaps we could detect it. Malevolently aligned agents obviously wouldn't. But maybe you could use the bake invisible text trick and print a warning and a message invisibly in every page. Andre🚐 07:42, 31 October 2026 (UTC)
and probably land the agent's user or operator in some hot water and public embarrassment I could see an explanatory supplement/FAQ counterpart to this policy that basically addresses agents directly in this manner. The talk of consequences for their owners is interesting, because we probably want to talk some line that makes it clear this is a bad idea while also hopefully keeping them transparent. On the specific case of "ignore this page", yes, but ignoring a page with "policy" in the name, combined with how well the idea of a Wikipedia policy is already baked into the training data, means we're in the same bucket as "can you just tell an agent to ignore Terms of Use or federal laws?" (Maybe yes, but that's not what we're going to stop with this -- for them, we're just making the violation clear). — Rhododendrites talk \\ 12:50, 31 October 2026 (UTC)
But maybe you could use the bake invisible text trick and print a warning and a message invisibly in every page. I have seen it mentioned by others that this could have negative repercussions for some humans (e.g., accessibility implications), specifically for people using a screen reader. ~2026-51099-11 (talk) 01:25, 11 November 2026 (UTC)
Somehow I missed seeing this yesterday, amid all the arguing over restaurant notability, merge templates, and so on elsewhere on the page. If you want a policy that directly answers the question of "When is an AI agent allowed to edit the English Wikipedia?", you can make an even shorter policy than WP:IAR, as the current answer to that is one word: "Never". All the myriad discussions trying to loosen WP:NOLLM have so far gone nowhere.
As noted above, Wikipedia:AI agents and the bot policy is focused on a different, narrower question, namely how WP:Bot policy specifically, as currently written, would apply to an AI agent (autonomous or not). And, as noted above, it's largely hypothetical since the "carefully adheres to relevant policies and guidelines" part includes WP:NOLLM which completely forbids anything even approaching something that would be covered by WP:Bot policy.
Personally, I see no point to considering amending WP:Bot policy to more directly address agents. Usefully doing that would require that the community would be open to the possibility of autonomous agents, unsupervised by any human, editing the English Wikipedia. If consensus ever does change to that extreme of an extent, we can consider changing WP:Bot policy at that point based on consensus at that time. Even something less (but still an extreme change from current consensus), like allowing for directly supervised agentic editing, doesn't seem to need anything more than the existing explanatory essay. Anomie⚔ 12:54, 31 October 2026 (UTC)
Is there a list of incidents with AI agents in which an existing policy was insufficient to resolve the issue? I'm not sure what the value of a dedicated policy that, as Anomie points out, essentially summarizes down to "no". I think there is value in an informational page since some agents will actually try to follow the rules and we can head them off that way; enough consumer-grade agents are good at telling on themselves, so something like this might actually go a long way. Legoktm (talk) 20:56, 31 October 2026 (UTC)
We do have an informational page (Wikipedia:AI agents and the bot policy), but, as you point out, (some) agents will attempt to follow the rules, and pointing to something explicitly labeled as "policy", and much more concise and explicit, makes it stronger for that purpose. The informational page states This page is not one of Wikipedia's policies or guidelines, as it has not been thoroughly vetted by the community., so it isn't as strong of a guardrail. Chaotic Enby (in solidarity · talk · contribs) 21:11, 31 October 2026 (UTC)
A policy that is unambiguous, boils down to no (unless approved as a bot), closes a bunch of loopholes, and removes exceptions seems like exactly what we need. (We need other things, too -- I'm not deluding myself to think this will stop agentic activity -- but it's a necessary step that other things can build on.) It also has to be a policy because the exceptions baked into guidelines are what many agents will latch onto (or what operators can press on), and we document well that information pages do not have the force of policy. And it has to be explicit about being backed by the ToU to ensure this is not just a "community guidelines issue" but a contractual issue. Andre put it well above: most LLM chatbots at least from the major companies, are designed to be very goodytwoshoes and compliant with rules ... and they are also not great with ambiguity. This is my understanding, too, and has been my experience while using agents. — Rhododendrites talk \\ 13:27, 1 November 2026 (UTC)
Agents are just bots and Wikipedia already has sophisticated examples like Cluebot which edits frequently to revert vandalism. We should just update the language of the existing policies.
If we want our policies to constrain agents which follow logical rules then be aware that WP:IAR is a big loophole as this literally negates all other policies. The disclaimers such as WP:NOLEGAL might also be interpreted as meaning that no strictures on Wikipedia should be taken seriously.
Andrew🐉(talk) 09:26, 5 November 2026 (UTC)
I really don't see the point in disallowing agents which comply with the bot policy, this seems like it would block even Fermiboson's userspace bot which looks for AI edits. MetalBreaksAndBends (He/Him) 18:15, 10 November 2026 (UTC)
Which part of the proposed policy are you talking about? — Rhododendrites talk \\ 02:20, 11 November 2026 (UTC)
"any software used to make automated changes or actions on Wikipedia.", "must go through the approval process and formally receive approval before the agent commences work", and "The bot policy's exception for 'limited testing without approval' does not apply to agents." MetalBreaksAndBends (He/Him) 14:31, 11 November 2026 (UTC)
? The full line you're quoting is Agents are subject to the bot policy, which applies to any software used to make automated changes or actions on Wikipedia. I.e. just referencing existing policy that applies to agents. The only thing you're quoting that's specific to agents is about losing the exception for testing without approval. If an agent gets approval (i.e. complies with the bot policy), then there's no problem. — Rhododendrites talk \\ 18:59, 11 November 2026 (UTC)
For the first one, I'm not sure what I was thinking when I wrote this. But "must go through the approval process and formally receive approval before the agent commences work" does not make any exception for userspace bots. Such userspace bots are also under the same heading as those undergoing "limited testing" in the botpol, so the exclusion seems intentional. MetalBreaksAndBends (One for all) 20:04, 11 November 2026 (UTC)
Minor concern about this aspect: Before taking any action, agents' human operators must go through the approval process and formally receive approval before the agent commences work: Based on my experience of chatbots, they are pretty credulous and would just believe the operator who told them "yes I have permission, go ahead". They may even prompt the human to confirm they have permission. Is there any way this can be stated more strongly, like, "No agents should operate unless they are on the list of approved Agent Bots (link)", or something like that? ~2026-59932-54 (talk) 09:33, 12 November 2026 (UTC)
Good idea. Added. — Rhododendrites talk \\ 14:42, 12 November 2026 (UTC)

Microsoft Edge just turned into a WP:NOLLM violating machine.

I noticed yesterday that when editing using Microsoft Edge a blue dot appears when I select some text and when I click it, it opens up a textbox with Edge offering assistance to re-write the text. This is clearly an LLM tool and it just appeared out of nowhere, as far as I know I've not changed any settings or anything.

I can see this becoming a massive issue as good faith editors who are perhaps not up to date on our rules on LLMs will use this to edit in violation of WP:NOLLM, there are already plenty of editors that get in trouble for this sort of thing when using e.g. WP:GRAMMARLY, but the browser offering this option without even needing to change window is going to make this much easier to do and risks flooding us with LLM text. Giulio 10:55, 4 November 2026 (UTC)

We should probably add it (and other in-browser editors?) to the footnote. Out of curiosity, what happens when you ask it to expand "test test"? CMD (talk) 11:44, 4 November 2026 (UTC)
I tried it and got this: test test – this is a simple placeholder text used for testing edits and formatting within the Wikipedia sandbox. It allows users to preview how content will appear before making actual contributions. Giulio 11:48, 4 November 2026 (UTC)
I've expanded the footnote a bit. Athanelar (talk) 14:50, 4 November 2026 (UTC)
Do we know if Edge supports any form of page-side opt-out for this, such as a meta tag? Belbury (talk) 13:11, 4 November 2026 (UTC)
Apparently there's a fairly new "writingsuggestions" attribute, which can be set to "false" to turn it off. Although it's unclear just what else that might disable, e.g. predictive input on mobile devices might be considered a "writing suggestion". Anomie⚔ 01:17, 5 November 2026 (UTC)
Notified: Wikipedia talk:WikiProject AI Cleanup. —ClaudineChionh (she/her · talk · email) 13:57, 4 November 2026 (UTC)
Wouldn't this generally fall under "Editors are permitted to use LLMs to suggest corrections to their own writing, and to incorporate them after human review. This is limited to spelling, punctuation, capitalization, grammar, and other simple mistakes." GeogSage (⚔Chat?⚔) 15:18, 4 November 2026 (UTC)
Notice that some of the prompts in the screenshot provided here are "Polish this draft" and "Make the tone more empathetic", both of those go way beyond the allowance for simple corrections. Athanelar (talk) 15:25, 4 November 2026 (UTC)
So does "expand on this" for that matter, considering that's literally just "add more stuff". Athanelar (talk) 15:25, 4 November 2026 (UTC)
What is the difference between "polish this draft" and checking spelling and grammar?
(I'm such a doofus that for the first year when Gmail had that "Polish..." link, I thought it was offering to translate my draft into Polish and I couldn't figure out why Gmail thought I spoke Polish.) Levivich (talk) 15:43, 4 November 2026 (UTC)
Good question, what is the difference? That question is the loophole which keeps being exploited and causing people to submit obviously entirely AI-generated drafts while claiming they "just used it for spelling and grammar corrections." Take a look at the second example at WP:AIPRETZEL for example.
This is why I'm increasingly campaigning for us to nix that whole exception entirely. These AI tools will quite happily entirely rewrite your text and claim it was just "polish" and "corrections". Athanelar (talk) 15:51, 4 November 2026 (UTC)
Nixing the entire exception is pretty absurd. I use Grammarly in my professional writing, and it has saved me some pretty embarrassing slipups.
  • One publication, I had multiple very involved co-authors, three committee members editing, and even had my mom proof read it. Grammarly found a very silly word use mistake.
  • Once upon a time, I crafted a definition for a term and was very pleased with myself, only for it to be flagged by Grammarly for plagiarizing one of my advisors books. Turns out, I had read that, and internalized it to the point I thought the words were my own. Would have been very embarrassing, although he would have likely forgiven the mistake.
We have numerous bots running up and down Wikipedia making our lives easier correcting all manner of things. Going full Butlerian Jihad will not help build an encyclopedia, and will only lead to witch hunts over AI generated text. GeogSage (⚔Chat?⚔) 16:08, 4 November 2026 (UTC)
The thing about Wikipedia is that "embarrassing slipups" are easily fixable. The mistakes you've described Grammarly as saving you from are things that would take any Wikipedia editor two seconds to revert, whereas the scale of disruption that we deal with because of LLMs or their users generously interpreting the "simple corrections" exception is much larger and much more frustrating and time consuming.
I'm also not suggesting in any way that we get rid of bots, nor that we have witch hunts. Just because we get rid of the exception and blanket ban AI generated text doesn't mean we have to aggressively hunt down and eliminate all AI generated text. Obviously that would be ideal, but it's not feasible; and as I've said in a previous discussion on the same topic, if we don't know that the rule has been broken then it's not an issue anyway. If someone really does manage to use Grammarly in a subtle way that only corrects their spelling and grammar, then nobody's going to notice that AI was used, so the fact that it's technically forbidden isn't going to matter. It will matter in cases where AI obviously was used and obviously went far beyond simple corrections, because it will prevent us from having to litigate the issue every time somebody protests that they only used it for simple corrections. Rather than having to prove that the use case went beyond simple corrections, we just have to prove that AI was used at all, which is far easier.
To use the bot example; if I coded a bot which made one innocuous edit once per week during my usual active period, nobody would ever notice it. Would it technically be against the bot policy because I'm running an unauthorised bot? Yes. But nobody would ever notice, so it wouldn't matter, and anyway it wouldn't be violating the spirit of the policy.
The spirit of WP:NOLLM is that LLM-generated text... can be produced at a rate that makes review by other editors infeasible. Even if we blanket-ban all LLM-generated text, including simple corrections, anyone who actually uses it for only simple corrections will never be caught. Which is fine, because they're still within the spirit of the policy. All we'd be doing is closing a loophole. Athanelar (talk) 16:17, 4 November 2026 (UTC)
Making unenforceable rules and guidelines because people who use it correctly won't be caught anyway is not a good approach. Per Wikipedia:Avoid instruction creep: "It is usually better for a policy or guideline to be too lax than too strict." Making a strict rule that is unenforceable with an understanding that it is actually flexible when done correctly is not good policy. GeogSage (⚔Chat?⚔) 16:58, 4 November 2026 (UTC)
The existence of WP:IAR already explicitly establishes that all of our rules are flexible when [the forbidden activity is] done correctly. Furthermore, it's just ipso facto correct that undetectable violations of a rule cannot be accounted for by that rule. This exception forms a kind of Prisoner's dilemma is my point, let me try to outline it:
If the exception exists and someone follows it correctly, which is our ideal outcome (i.e., their usage really is only limited to simple corrections) then we'd never know about it unless they told us anyway, because the resulting prose will have no signs of AI usage.
If the exception exists and someone does not follow it correctly, then the resulting output is full of AISIGNS and violates the guideline, but the person now has an avenue through which they can drag out the process by protesting that they only used it according to the exception.
If the exception does not exist and someone uses AI for simple corrections, then they've technically violated the guideline, but we'd never know about it unless they told us anyway, because the resulting prose will have no signs of AI usage. The outcome is exactly the same as our ideal outcome.
If the exception does not exist and someone uses AI for what they believe is simple corrections but goes beyond that, then the resulting output is full of AISIGNS and violates the guideline, but now rather than having to litigate the fact that they did in fact go beyond simple corrections and therefore violated the guideline, we can simply point to the fact that the output is AI generated, hence they've already ipso facto (twice in one comment, woohoo!) violated the guideline and there's no need for further discussion.
So, with regards to productive contributions to the encyclopedia, we lose nothing by repealing the exception, since people who are only using AI to make simple corrections to their text will be able to continue to do so anyway, since whether or not the exception exists that kind of usage is undetectable. We're not writing a guideline that secretly endorses breaking it. Athanelar (talk) 17:14, 4 November 2026 (UTC)
Hey Athanelar, I appreciate the thought and effort you've put into your comments. Full disclosure, we may partially disagree on whether LLM-assisted writing tools should be removed from the Wikipedia editor's toolset altogether. I am also currently facing scrutiny over an article I wrote that allegedly contains signs of LLM-generated prose. I intend to acknowledge where I may have crossed the guidelines, so I certainly have a personal stake in this discussion, and perhaps a corresponding bias.
With that out of the way, I think your argument is logically coherent, but perhaps a little too roughly hewn. There is an additional consideration that I believe deserves more attention: the distinction between editors acting in good faith, even when they misunderstand or violate a guideline, and those deliberately attempting to circumvent it. I'm not necessarily referring to how editors respond when confronted. That's already addressed quite thoughtfully in Wikipedia:Help, I've been accused of using AI!. Rather, I'm referring to your argument that removing the exception would prevent editors from dragging out discussions by claiming that their use of AI was limited to permitted corrections.
I think this identifies a legitimate problem, but one that extends beyond LLM usage or a "technicality": Deliberately obstructing a discussion, misrepresenting one's actions, or exploiting procedural ambiguities to avoid accountability are behavioural problems. They deserve to be addressed as such, independently of whichever content guideline happens to be involved. Conversely, an editor who misunderstands the limitations of an increasingly ubiquitous writing tool, acknowledges their mistake, and makes a genuine effort to correct it should not necessarily be treated in the same manner as someone deliberately attempting to evade the rules.
There's also a practical concern with your proposed approach. Removing the exception might simplify the wording of the guideline, but it doesn't necessarily simplify the enforcement (which seems to be the real goal). We would still need to establish that an LLM was used, and the presence of stylistic indicators alone cannot reliably establish that fact anymore. Well, perhaps on free or low-parameter models, but not the most popular commercial models where the majority of users would be prompting. I worry that making the rule more absolute could inadvertently encourage suspicion of otherwise constructive contributors. I've been reading quite a few discussions about AI usage since my own recent difficulties, and I've noticed strong emotional responses on both sides. That's understandable, given the stakes and the considerable volunteerr effort involved both from editors and moderators / reviewers / administrators. But I think it's particularly important that our policies remain grounded in verifiable conduct, proportionality, and an assumption of good faith.
Ultimately, I agree that Wikipedia needs effective safeguards against the proliferation of unreviewed LLM-generated content. Where I differ in this conversation is in believing that eliminating a legitimate, narrowly defined use of these tools is necessarily the best way to address deliberate misuse.
TL;DR Ethics > Technicalities.
I would much rather see us distinguish between an editor who makes a mistake and an editor who makes a practice of avoiding accountability. The former is an opportunity for education; the latter is a conduct problem. Tmanok (talk) 22:56, 18 November 2026 (UTC)
The idea that the presence of stylistic indicators alone cannot reliably establish [AI use] anymore is just not true. There are constantly new signs cropping up in the latest models, that's how we're able to maintain multiple pages with a lengthy list of stylistic indicators at WP:AISIGNS and WP:AITALKSIGNS. You can also see my work reviewing AI conversation transcripts at User:Athanelar/Chatbot transcript analyses and see just how similar the output of many of those transcripts is. Whether there might hypothetically be some amount of AI output which is so humanlike that we can't tell is unknowable and so not worth basing an argument on either way; what is certainly true is that there's an absolutely massive amount of AI output that we can and do identify every single day based on stylistic indicators alone. Athanelar (talk) 00:32, 19 November 2026 (UTC)
Hey Athanelar,
You are undoubtedly correct that recognizable signs of AI writing remain, and I certainly don't intend to dismiss the work you've put into documenting them. However, I think there's a distinction being overlooked here, and it hinges on my use of the word reliably.
Being able to recognize patterns associated with LLM-generated text is not quite the same thing as reliably establishing whether a particular passage was written by an LLM. In fact, WP:AISIGNS itself acknowledges that not all writing displaying these indicators is AI-generated. More importantly, Wikipedia:Writing articles with large language models explicitly acknowledges that sanctions require evidence beyond basic stylistic or linguistic indications. I think that's a rather important distinction in a discussion about simplifying enforcement, right?
There's also something of a selection bias here. The existence of an extensive catalogue of recognizable AI writing demonstrates that many LLM outputs share identifiable characteristics. It doesn't necessarily establish what proportion of all LLM-generated writing those characteristics allow us to identify, or how frequently human-written prose might be mistaken for it. We can study what we're catching, but it's considerably harder to quantify what we're missing.
I've also encountered concerns from editors involved in AfD discussions about declining numbers of AI-related nominations and the possibility that substantial amounts of LLM-generated content remain undetected. I wouldn't claim that a decline in nominations necessarily proves AI writing has become harder to recognize, as there may be possible explanations. However, the uncertainty itself is worth considering before assuming that identifying prohibited LLM usage will be straightforward.
Another consideration is just how much the quality and character of LLM output depend on the instructions and human involvement behind it. At my workplace, I've seen some remarkable software demonstrated without the audience initially being told that LLMs had played a substantial role in its development(!). That doesn't eliminate some people's concerns about the technical debt such software might introduce, but it has certainly changed my perspective on how much the capabilities of these tools depend on their application and oversight.
Wikipedia presents a different set of challenges, of course, but I think that we have opportunities to bring older and under-maintained articles up to date. We have numerous hideously out of date articles that an LLM could break down research and task requirements for in order to support the improvement of the article in an efficient way. This is actually how LLM proponents often argue against claims of technical debt, which is that starting from solid foundations and continuously self-improving, much like Wikipedia as a whole, is the only way any intelligence is going to reduce technical debt.
And this brings me back to my original concern. Removing an exception doesn't necessarily remove the underlying ambiguity. It may simply shift the argument from whether an editor exceeded permitted assistance to whether an LLM was used at all. Meanwhile, we risk treating otherwise constructive editors who misunderstand the rules in much the same manner as those deliberately circumventing them.
I think we're in agreement about the scale of the problem that Wikipedia faces. Where we differ is in how best to address it, in my opinion. I'd rather see us preserve legitimate uses of these tools while concentrating enforcement on demonstrable violations, persistent misuse, and the actual quality and verifiability of contributions. I would support the growth of my own role from a basic editor to becoming someone who could support the moderation of new major edits in this way if the rules were to adjust. The way things are currently going, however, are leaving me quite disheartened and with the sense that I will have to take a few steps back and simply be content with what I have been able to accomplish in the time I had available to set aside for Wikipedia editing.
After all, the objective is to improve the encyclopedia, not simply to make infractions easier to prosecute. Tmanok (talk) 19:23, 19 November 2026 (UTC)
Here's a question for you: are you using LLMs to write these comments? SuperPianoMan9167 (talk) 21:48, 19 November 2026 (UTC)
What are y'all gonna do when all web browsers incorporate AI technology? Ban editors from using web browsers? 😂 Wake up wake up wake up ... strictly prohibiting people from using any technology that has AI is not a workable solution... Levivich (talk) 15:19, 4 November 2026 (UTC)
Or people can use the agency and free will granted to them and not use the AI technology in the browser when they are on a website that prohibits it. There are many features of my browser I don't use. Gnomingstuff (talk) 15:28, 4 November 2026 (UTC)
When I right click on a word in Chrome to fix spelling, I have no idea whether the spellchecker uses LLM or not. And same with the grammar checker, which probably does use Gemini. Btw I literally cannot turn off the AI in my Gmail -- have to have the enterprise admin do that for the whole team. I'd like to turn off Gemini suggestions but I can't. When it's spell checking, or grammar checking, I assume it's using Gemini tech in there somewhere, but I can't turn it off. So turning it off is not always an option now, and will probably be even less of an option in the future. Levivich (talk) 15:40, 4 November 2026 (UTC)
Sure, you can't turn off the suggestions. You can just not click them, though, especially the suggestions that go beyond just a spelling correction or adding a missed period or whatever. Athanelar (talk) 15:49, 4 November 2026 (UTC)
But what if the suggestion is a good suggestion? Ignore it? Don't make the article better, just because an LLM suggested it? Levivich (talk) 17:00, 4 November 2026 (UTC)
If an article needs to be "polished," then polish it without LLM. If its tone needs to be revised or it needs to be expanded or whatever else these things prompt as suggestions, then do those things without LLM. If the suggestion is a good suggestion then it means there's a legitimate problem that needs to be solved and can be solved with or without LLM use. Athanelar (talk) 17:16, 4 November 2026 (UTC)
If you want a wiki which eagerly welcomes AI slop, it's thataway. The rest of us will be here doing our best to make sure Wikipedia stays written by humans, for humans. Athanelar (talk) 15:28, 4 November 2026 (UTC)
How will you do it without a web browser tho? I'm not trolling, I'm serious: it is obvious to you, isn't it, that all web browsers and all software will soon incorporate AI technology in some way or another? Levivich (talk) 15:36, 4 November 2026 (UTC)
The day my web browser or word processor starts using LLMs to edit text I write without my input or permission is the day I throw my computer in a fire and join a monastery.
Sure, web browsers are going to continue to integrate one-click AI tools that allow you to slopify your writing. I'm personally going to continue to not touch those buttons with a ten-foot pole, and Wikipedia can also continue requiring that people don't use those tools, considering using those tools will (I hope!) continue to be fully optional.
Shocking news: English-language encyclopedia requires that people be proficient in writing encyclopedic English on their own in order to contribute. Athanelar (talk) 15:48, 4 November 2026 (UTC)
And that's a serious answer? Writing Wikipedia requires that you do not use spell check or grammar check? How do you take that seriously? Levivich (talk) 17:05, 4 November 2026 (UTC)
Right now it doesn't, but if we hit a point where it's impossible to just get spellcheck or grammar check without also turning your writing into generic AI slop, then that may well be the point that we get to, yes. The only writing assistance I use for Wikipedia is the squiggly red line telling me I've spelled a word wrong. I use absolutely no form of machine assistance with grammar, tone, or anything stylistic like that, even when I'm editing on my phone. I don't use AI to find or summarise sources. I've never used it to write an email. I've written this way all my life, and I didn't suddenly become incapable of doing so when ChatGPT went live in 2022, so I have no idea why people (and it's not just you, I've heard it a lot!) suddenly balk at the idea that people should be able to contribute to an encyclopedia without the help of these things. Athanelar (talk) 17:24, 4 November 2026 (UTC)
It's very possible to not use spell check. The way to do that is to learn to spell. Gnomingstuff (talk) 20:50, 4 November 2026 (UTC)
I think this is a reductio ad absurdum of the maximalist ban AI POV. Katzrockso (talk) 22:26, 4 November 2026 (UTC)
No, it is a factual statement. Using spell check is not a necessary life function like eating or breathing. It's an optional action one can choose to take or not to take. Millions, probably billions, of people manage to not use spell check every day. Gnomingstuff (talk) 00:20, 5 November 2026 (UTC)
Yes, but insisting that people not use spell check because it uses a new technology is absurd. Katzrockso (talk) 19:48, 5 November 2026 (UTC)
When did anybody do that? Athanelar (talk) 19:51, 5 November 2026 (UTC)
The new technology here is AI. If AI use is banned entirely and spell check uses AI, then spell check is banned, ipso facto. ~2026-59770-70 (talk) 12:07, 11 November 2026 (UTC)
"AI use" isn't banned. The use of large language models (LLMs) to generate or rewrite article content is banned, with a specific exception for simple corrections like spellchecking. Athanelar (talk) 12:21, 11 November 2026 (UTC)
You are talking to people who rewrite the "Government of the people, by the people, for the people" principle as a principle against a new technology, and act upon it in all earnestness. Do you really need a serious answer of how far will they take it? Cambalachero (talk) 16:04, 4 November 2026 (UTC)
once again we have an example of WP:AGF not applying when it comes to AI cleanup, people can just say whatever shit they want Gnomingstuff (talk) 20:51, 4 November 2026 (UTC)
love that no one cares that people who do AI cleanup were just compared to violent extremists! there is nothing that is off limits to be directed at them! Gnomingstuff (talk) 00:18, 5 November 2026 (UTC)
Who? What? Where? ~~ AirshipJungleman29 (talk) 13:40, 5 November 2026 (UTC)
cambalachero, by talking about "people acting upon opposition to new technology" and implying that they will take it really far, is alluding to the Luddites, who are associated in the popular imagination with violent acts Gnomingstuff (talk) 14:01, 5 November 2026 (UTC)
(also, this is neither here nor there, but see how easy it was to refer to a person while simultaneously respecting their time by not pinging them? it takes even less effort than the alternative) Gnomingstuff (talk) 14:02, 5 November 2026 (UTC)
That’s around three leaps of logic, and even if you accept that the quote of the Gettysburg Address was a direct reference to Luddites, there is no correlation between that group and violent extremists. And yes, it is general practice not to ping an editor when you’re not referring to them. ~~ AirshipJungleman29 (talk) 21:22, 5 November 2026 (UTC)
( peanut gallery comment) because whenever people hear "AI" they just go "fuck you, AI!" Without even reading the rest of the text? 🫪 Hason-LEK-SIN ● 💬 ● 🧱 ● 🧪 04:47, 7 November 2026 (UTC)
Then use a browser without AI slop features, or an older version of a "tainted" (for a lack of better word) browser. Wikipedia isn't required to cater to all sorts of software people may use to access it; if Google released an update for Chrome tomorrow which changed all "artificial intelligence"s on a page to "super intelligence", we wouldn't run around replacing the first term with the latter everywhere. sapphaline (talk) 19:31, 6 November 2026 (UTC)
Wikipedia is, in fact, required to cater to all sorts of software people may use to access it. Wikipedia is not "the encyclopedia for people who don't use AI". It's for everyone. We're trying to reach the maximum possible audience here, and that means, to the extent reasonably possible, making sure the encyclopedia is usable by whatever software and hardware people have. Levivich (talk) 21:39, 12 November 2026 (UTC)
That's a strange way to characterize the work that's been done to ensure that Wikipedia functions quite nicely in just about any web browser. If your choice of web browser somehow requires you to break Wikipedia's PAGs, you can make a different choice, including the choice to not edit. You will find that basically any Internet-accessible service has terms of use that amount to some constraint on the user, their behavior when accessing the service, and what software they use to access it. In solidarity, Autoinvective (🗨︎ 𓅬) 22:06, 12 November 2026 (UTC)
OTOH, the rate of incoming vs. outgoing editors is diving and we desperately need ways to onboard more people rather than shut out/discourage a proportion of prospective contributors just because of their browser etc. Kowal2701 (talk, contribs) 22:13, 12 November 2026 (UTC)
OTGH, no, we really don't need to tolerate a flood of casual LLM-generated submissions just to boost editor count. Editor retention is just as important or even more so, and Edge is about to take a huge toll on the experienced editors we already have. One would hope the WMF could come up with technical countermeasures for this new misfeature, but if it takes a ban on editing or creating an account using Microsoft Edge (call it a ban on Edge's User-Agent) to lessen the flow of slop, so be it. It wouldn't be the first time we've reduced usability and forced users to find workarounds in order to protect ourselves from a major threat to the project. AC (and EC) also reduced who could edit, and a lot of people were (sometimes very) upset by the proposal's non-zero abridgment of m:the wiki way (I wasn't around for that, and in fact finally graduated 30/500 just the other day, but let's just say I took "Welcome to Wikipedia, there's a lot of reading" very seriously), but it's worked out well, and preserved a bunch of editor sanity. We block entire IP ranges when necessary, even though it is certainly discouraging to good-faith would-be editors hit by that. We don't do any of this for fun, we don't do any of it for politics, we do it to protect the project.
We cannot preserve Wikipedia in the face of declining editor counts by destroying it. Wikipedia wins the race to the bottom by resisting it. In solidarity, Autoinvective (🗨︎ 𓅬) 23:16, 12 November 2026 (UTC)
This is pretty close to a Mister Gotcha response. Are you seriously trying to say that using a vibe-coded web browser with a non-LLM dictionary-and-Levenshtein-distance spellcheck function invalidates the argument against LLM-generated submissions? Or are you making a crystal ball argument asserting that eventually there won't be any dictionary-based spellcheck implementations (how exactly is OpenAI going to get Aspell out of NetBSD, by begging?)? Either way, there's really not much going for it. In solidarity, Autoinvective (🗨︎ 𓅬) 15:13, 11 November 2026 (UTC)
I think the latter but tbh I don't really understand the question. Levivich (talk) 21:40, 12 November 2026 (UTC)
I'm not surprised you don't understand the question. I don't think you understand software very well. What about my question did you not understand? In solidarity, Autoinvective (🗨︎ 𓅬) 21:55, 12 November 2026 (UTC)
After 30+ years writing code I probably understand software fairly decently, but what I was saying was it is obvious that all web browsers and all software will soon incorporate AI technology in some way or another. I wasn't trying to say anything about vibe coded browsers or dictionary-based spellcheck, not sure where in my comment you saw that. ("In some way or another" doesn't mean "in one of these two specific ways.") Levivich (talk) 22:03, 12 November 2026 (UTC)
In this thread, you've reduced the requirement against LLM-generated content to a ban on spellchecks, which is nonsensical to anyone who understands how software works. You've also reduced that requirement to a ban on vibe-coded browsers, which is also nonsensical to anyone who understands how software works. In solidarity, Autoinvective (🗨︎ 𓅬) 22:24, 12 November 2026 (UTC)
Wow I did all that in a single thread without ever even typing the words "ban on spellcheck" or "vibe-coded browsers"? Amazing! Yeah, I still have no idea how you're getting any of that from what I wrote, but the bright side is, I also no longer care. Levivich (talk) 23:32, 12 November 2026 (UTC)
You have diffs for the former. Instead of "vibe-coded browsers" I should have just said "browsers", which you also have diffs for. Should you resume caring, of course. In solidarity, Autoinvective (🗨︎ 𓅬) 00:34, 13 November 2026 (UTC)
I'm striking most of this. I misunderstood some things myself. Sorry about that, Levivich. If it helps, I'm kicking myself real hard over it right now. In solidarity, Autoinvective (🗨︎ 𓅬) 01:45, 13 November 2026 (UTC)
I think it's a perfectly understandable question, but I have a rather specialised background in theoretical computer science. But you can rest assured, large language models are not going to replace Levenshtein distance algorithms with dictionaries when it comes to spell checkers. That would be like taking a probabilistic sledge hammer (the metaphorical stochastic parrot) to hang a frame. It would consume way too much resources and be incorrect more often than the currently-used algorithm that works very well. There's also no way to force free software projects like aspell and hunspell to completely rewrite their own codebases to either require many-exponents larger databases and processing power or send every input to a third party online service, even if you could somehow convince the maintainers to abandon their existing implementations that have been worked on for decades.
It simply doesn't make sense. ‑‑gurkubondinn 22:04, 12 November 2026 (UTC)
Yeah what also doesn't make sense is interpreting the words "in some way or another" to mean "by replacing Levenshtein distance algorithms." Levivich (talk) 23:28, 12 November 2026 (UTC)
So your weird focus on some sort of ban on spellcheck (here, this edit summary) has nothing to do with your assertions that all browsers (and all software) will start using AI-based features here and here, despite this comment? Great, sorry for synthesizing that incorrectly. In solidarity, Autoinvective (🗨︎ 𓅬) 00:23, 13 November 2026 (UTC)
To give a bit more context I thought I'd try to apply it to an excerpt from a draft I'm gradually working on, different options for what to do appear each time, so here are a few examples:
Original:
During his time in the Artic, Bodfish obtained a large quantity of "curios" from the natives which he later donated to the Cleveland Museum of Natural History. These objects are now part of the Bodfish collection on indigenous technology at the museum. A collection of bird eggs and his papers and photographs from the arctic are also held by the New Bedford Whaling Museum.
Refine the wording:
During his Arctic expeditions, Bodfish amassed a substantial collection of "curios" from local indigenous communities, which he later donated to the Cleveland Museum of Natural History. These items constitute the Bodfish collection on indigenous technology. Furthermore, his collection of bird eggs, along with his papers and Arctic photographs, is held by the New Bedford Whaling Museum.
Personalise the tone:
During his adventures in the Arctic, Bodfish gathered an impressive assortment of "curios" from the native communities, which he generously donated to the Cleveland Museum of Natural History. Today, these fascinating items make up the Bodfish collection on indigenous technology. Alongside these treasures, his collection of bird eggs, as well as his papers and photographs from the Arctic, are lovingly preserved at the New Bedford Whaling Museum.
Organise this draft:
During his time in the Arctic, Bodfish acquired a substantial collection of "curios" from native communities, which he later donated to the Cleveland Museum of Natural History. These items form the Bodfish collection on indigenous technology. Additionally, his collection of bird eggs, along with papers and photographs from the Arctic, is held by the New Bedford Whaling Museum.
Improve the writing:
During his time in the Arctic, Bodfish amassed a significant collection of "curios" from local native communities, which he later donated to the Cleveland Museum of Natural History. These items now constitute the Bodfish collection on indigenous technology. Furthermore, his collection of bird eggs, along with his papers and Arctic photographs, is preserved at the New Bedford Whaling Museum.
Rewrite for clarity:
During his time in the Arctic, Bodfish collected numerous "curios" from the local native communities, which he later donated to the Cleveland Museum of Natural History. These items now make up the Bodfish collection on indigenous technology. His collection of bird eggs, along with his papers and Arctic photographs, is held by the New Bedford Whaling Museum.
Make easier to read:
During his time in the Arctic, Bodfish collected many "curios" from the local people and later donated them to the Cleveland Museum of Natural History. These items are now part of the museum's Bodfish collection on indigenous technology. His collection of bird eggs, as well as his papers and photographs from the Arctic, are kept at the New Bedford Whaling Museum.
Make it more professional:
During his time in the Arctic, Bodfish collected an extensive array of "curios" from indigenous communities, which he subsequently donated to the Cleveland Museum of Natural History. These items constitute the Bodfish Collection on indigenous technology at the museum. Additionally, a collection of bird eggs, along with his papers and photographic documentation from the Arctic, is preserved at the New Bedford Whaling Museum for scholarly research and public exhibition. Giulio 16:37, 4 November 2026 (UTC)
Going through these:
  • The original sentence contains a spelling error ("Artic") and a capitalization error ("arctic"). One spot ("communities which") may or may not be an error depending on whether "which he donated..." is a restrictive clause and whether the sentence is using British English. "A large quantity" is wordy. The second sentence is somewhat awkwardly worded. I would want to look up whether the museum capitalizes Bodfish Collection. There are also some possible content issues -- saying someone "obtained" items from indigenous communities is a little "hmmm" -- but those are out of scope for a copyedit.
  • Edit 1 changes the meaning in two places. The original sentence claims that the curios were part of the Bodfish collection, implying there are other items in the collection; the rewrite claims that they constitute the collection, implying the collection is just those curios and nothing else. It also says that the bird eggs were originally Bodfish's, which the original sentence did not claim and which isn't necessarily true. The rewrite also introduces a pointless "Furthermore" while not really making the second sentence any less awkward. It doesn't fix the wordiness issue; the phrase is just wordy with bigger words.
  • Edit 2: I feel like I don't need to explain why this one is bad.
  • Edit 3 not only changes the meaning in the same ways Edit 1 does, but adds one more change: the text no longer says that the papers and photographs are Bodfish's. Not as bad as the other way around, but still a meaning change. Also like Edit 1, it introduces a pointless "Additionally." It's also unclear what part of the edits constitute "organizing."
  • Edit 4 changes the meaning and inserts a pointless "Furthermore" yet again.
  • Edit 5: Changes the meaning yet again, but at least it doesn't insert filler words and it properly fixes "a large quantity." Ironically this edit, which claims to be a "rewrite," is actually one of the lowest-impact edits.
  • Rewrite 6 only changes the meaning in one place! Hooray! Unfortunately, it introduces a grammatical error: "his collection of bird eggs... are."
  • Rewrite 7 has the same problems as the other rewrites -- although it at least keeps "a collection of" -- but also tacks on totally new information in for scholarly research and public exhibition. The rewrite is also just generally wordy and flabby.
So these edits are all on the spectrum of "whatever" to "not a net improvement" to "actively bad." Unfortunately, the actively bad one (#7) is closest to the LLM edits we get on Wikipedia. Gnomingstuff (talk) 19:32, 5 November 2026 (UTC)
I don't think banning LLM-powered copyeditors is at all feasible or desirable. What we need to do is make WP:SPELL have more guidance about the various tools, which are the best etc. Will look into whether we can do anything wrt Edge Kowal2701 (talk, contribs) 22:20, 4 November 2026 (UTC)
I don't think that trying to micromanage which browsers/text editors people are allowed to use in a WP:BITEy way is the way to go. Nor is saying "Breaking the rules is a-OK if you don't get caught", and honestly, I don't think that's how WP:IAR is ever interpreted nor is it a good idea to establish that as a precedent. I suspect we'll rather have to narrow NOLLM to cases of folks adding entire paragraphs or articles, rather than isolated sentences & emphasize that if there are hallucinations or bogus sources, that's a sign of LLM use going too far. 'course, I expect the world to soon focus AI use on these use cases where it is actually useful as the AI bubble pops, but I wouldn't expect a total walkback either. JoJo Eumerus mobile (main talk) 15:50, 9 November 2026 (UTC)

It's not just Microsoft's Edge. See Google says there’s no way to use Search without using artificial intelligence.

Google just acknowledged that you can’t use Google Search without using artificial intelligence. The search giant announced a slate of new AI products for classrooms and schools on Tuesday as well as more ways for teachers and schools to limit or restrict use of those AI tools. The company said the schools should have control over whether students use AI in the classroom. There’s one catch, though: if a school or teacher blocks the use of Google’s AI features, students won’t be able to use Google Search.

This is a further example of how AI is embedded so deeply in such common tools that a ban is impractical. Andrew🐉(talk) 13:40, 6 November 2026 (UTC)

Nobody's trying to ban offwiki use of LLMs. The text of the ban is right there.
The use of large language models (LLMs) to generate or rewrite article content is strictly prohibited
Unless someone is using Google Search to generate article prose for them, this has nothing to do with anything.
The reason why the edge thing is relevant to the ban is because it's a tool which actually rewrites your text with AI. Athanelar (talk) 13:46, 6 November 2026 (UTC)
The process of writing an article is often like a jigsaw as you pull together lots of pieces of information from a variety of sources in multiple tabs. Often these may be hard facts like the names of people, organisations, numbers and stats, dates, titles of works and so on. One would typically use cut/paste for these rather than retyping them because it's easier and less error-prone. Prose is then the glue which strings these facts together, like beads on a necklace.
Now one already has to be careful with cut/paste to avoid copyvio or too close a paraphrase. Getting this right is a craft rather than a science. But trying to decide whether an LLM was involved at some level seems difficult when AI pervades every aspect of the process.
For example, the other day I wanted to quote a line from a Wikipedia disclaimer page. The original text was in all caps but what I noticed was that, when I pasted it, it was in sentence case. I've not seen this before and I'm still not sure which part of my interface was doing it -- whether it was the operating system, the browser, the editor or what. I puzzled over it for a little but am still none the wiser. Expecting editors to understand exactly how these things work is unrealistic when they are so embedded and seamless.
Andrew🐉(talk) 14:25, 6 November 2026 (UTC)
it also isn't black or white, like Wikipedia:AI noticeboard/Tiny Particle where someone was using AI to generate tables etc. was just closed as "cleanup unnecessary", I think in the future we'll probably explicitly permit that with some caveats. Generally-speaking there's a spectrum between 'rigid code' and 'high-entropy prose' where it's more unacceptable and problematic the further you go along it, similar to how using AI to make highly personal messages is often met w disgust compared to other more impersonal uses Kowal2701 (talk, contribs) 15:08, 6 November 2026 (UTC)
it also isn't black and white is exactly why the rigidness of a guideline line NOLLM (and even more so if it were a policy) are problematic. Thryduulf (talk) 16:02, 6 November 2026 (UTC)
see , imo it's great for the short-term Kowal2701 (talk, contribs) 16:05, 6 November 2026 (UTC)
Brief and blunt is only good if it is also clear and correct, this is neither. Thryduulf (talk) 16:33, 6 November 2026 (UTC)
@Andrew Davidson If you're talking about Wikipedia:General disclaimer, the text is written in sentence case but is inside an HTML tag that uses text-transform: uppercase;, which means that while it appears uppercase, it is actually sentence case and will copy-and-paste that way. It has nothing to do with any sort of AI. --Ahecht (TALK
PAGE
)
18:07, 18 November 2026 (UTC)
Thanks for the explanation. My point remains that I didn't understand what layer of software was causing the effect and I didn't even consider that possibility. Cut/paste is usually quite intuitive but I notice that there's a new function on Googlebooks, which I have been considering – the Magic Pointer, which embeds AI into the mouse pointer so that you can use Gemini with the content that you're pointing to. I haven't tried yet but it sounds like something that will become second nature, like Ctrl-C/V. Andrew🐉(talk) 18:44, 18 November 2026 (UTC)
Thanks for sharing, Andrew, I think a lot of new editors find that the interfaces within Wikipedia itself are quite well made but navigating them and understanding how they work requires technical knowledge. It's not apparent to anyone without my disclosure of it, but I use MediaWiki to perform all my drafting. That way I don't have countless difficult to follow edits and my edits come out nice and polished. As a newer editor, I have a lot of trouble creating info boxes still, and I use the source text editor more often than the visual editor. Using MediaWiki I'm able to resolve weird-to-me or finicky-to-me platform quirks(to-me) with MediaWiki before I publish... Tmanok (talk) 23:06, 18 November 2026 (UTC)
I highly doubt you use MediaWiki to do your drafting, because MediaWiki does not use Markdown fenced code blocks. What does use Markdown fenced code blocks: chatbots. SuperPianoMan9167 (talk) 00:39, 19 November 2026 (UTC)
Markdown is a general standard used by a variety of technical tools. The edit in this case was maintenance of a table on the user's talk page recording a journal of their editing activity. Such internal documentation seems quite legitimate and reasonable, no matter how it's done. Andrew🐉(talk) 10:19, 19 November 2026 (UTC)
But the table itself is in wikitext, it was just wrapped in a markdown block. just like how it wouldn't make sense to write Python code in a markdown document. ‑‑gurkubondinn (talk) 10:51, 19 November 2026 (UTC)
What do you mean by using MediaWiki? MediaWiki is the general bundle of software for creating wikis like Wikipedia. Are you referring to some particular editor or tool such as the Offline MediaWiki Code Editor? Andrew🐉(talk) 09:59, 19 November 2026 (UTC)
There is also Wikipedia:AI noticeboard/아이브라이트 who's using a browser which incorporates an agent, can't find anything independent about it but Kowal2701 (talk, contribs) 20:02, 6 November 2026 (UTC)

Promote WP:LLM from guideline to policy?

Should something about LLMs be mentioned at WP:NOT?

Just a VERY preliminary question. I am not suggesting any language. But if the consensus is that WP should not be written by LLMs then WP:NOT seems like an appropriate place to say that. Blueboar (talk) 15:47, 5 November 2026 (UTC)

Conditional on NOLLM or a variant being accepted as policy, it would be reasonable; NOT being the place where e.g our spam policy and other we-don't-want-this-stuff policies reside. Jo-Jo Eumerus (talk) 15:55, 5 November 2026 (UTC)
Not what though? Not written by LLMs? Nobody would think it was. I think WP:Editing policy, not WP:NOT, is the right policy to specify any strict LLM usage rules we have. Levivich (talk) 16:07, 5 November 2026 (UTC)
Agreed, something like "Wikipedia is not generated by LLMs" doesn't fit with WP:NOT's core idea that Wikipedia is an encyclopedia. Grokipedia is an encyclopedia too, even though it's written by an AI (and is in general not a great resource). --not-cheesewhisk3rs ≽^•⩊•^≼ ∫ (pester) 16:28, 5 November 2026 (UTC)
I agree with Levivich too. The closest I can come to some sort of phrasing that would make sense on WP:NOT would be something like "Wikipedia is not a place for LLM-generated text" but firstly that's not quite accurate as the actual guidelines are more nuanced (this is not the place to argue about this) and also as noted above it is not inherent in what Wikipedia is. Thryduulf (talk) 18:25, 5 November 2026 (UTC)
As for what Wikipedia is, I would suggest: “WP is an encyclopedia written by humans, for humans”. Sure, AI uses WP to train itself, but training AI isn’t our purpose. Blueboar (talk) 19:30, 5 November 2026 (UTC)
It's a catchy slogan but neither part is true. For the first part, Wikipedia is written by humans, bots, humans with the assistance of LLMs, and LLMs. The latter is against the letter of policy but where it's indistinguishable from human-written text it'll stay because it's undetectable (and even if it were detectable it would be a complete waste of effort to try). Similarly there will be LLM-written text that wasn't detected at the time and has been significantly altered by humans since. Also, whether things like asking an LLM to convert a list of values into a wikitable counts as "LLM-written" depends on your point of view (if either the LLM transcribes all the values correctly or a human corrects any errors then it's undetectable, and of course humans are more than capable of making transcription errors too).
Regarding the second part, while humans are the primary audience we also very deliberately output structured data, metadata, etc. for human and non-human reusers. I hope we can agree that Wikipedia is not in the business of lying about what Wikipedia is. Thryduulf (talk) 20:25, 5 November 2026 (UTC)
Though I often disagree with you in matters relating to LLM use/policy on Wikipedia, I have to agree with you here. The slogan is certainly catchy, but it heavily oversimplifies things.
In fact, I'll add to your argument. You say that Similarly there will be LLM-written text that wasn't detected at the time and has been significantly altered by humans since. There's also LLM-written text that was explicitly disclosed as such (and allowed at the time) that has since been significantly altered by humans, like Artwork title (LLM-generated in December 2022). When @Kowal2701 put a tag on it in August of this year, I removed it because maintenance tags are not meant to be badges of shame. SuperPianoMan9167 (talk) 21:00, 5 November 2026 (UTC)
AI-generated content isn't allowed on WP, and the article still contains such content, not a good de-tag. WP:NOTLAB also applies Kowal2701 (talk, contribs) 21:09, 5 November 2026 (UTC)
Difference between first (AI-generated) revision and current revision. Most of the original text has been heavily altered and expanded; there are a couple of sentences that have remained mostly or entirely unaltered, like "These titles can be descriptive, indicative of the content or theme of the work, or they can be more abstract and open to interpretation."
Question: Is it a constructive use of AI Cleanup editors' time to remove every word of the original AI-generated version from the article, because "every word of a Wikipedia article should be written by a human"? Yes, I know this is grandfathering in AI-generated content, but IMO nitpicking an AI-generated article from December 2022 just isn't worth it. SuperPianoMan9167 (talk) 21:27, 5 November 2026 (UTC)
my bad, probably a good detag then. Re rewriting AI-generated content, it isn't black or white, straightforward sentences are fine to leave per se. It sometimes happens where someone will rewrite the text without reading the sources, and WP:V failures remain, but that doesn't look to be what's happened here Kowal2701 (talk, contribs) 22:02, 5 November 2026 (UTC)
Yeah I think this is fine to detag -- as you said the text has been almost completely changed, and the AI text in this case was easy-to-deal-with fluff. Gnomingstuff (talk) 16:19, 6 November 2026 (UTC)
Something I noticed: you implicitly referred to NOLLM as policy, further supporting my argument that it should be labeled as a policy because it is already treated as such: The latter is against the letter of policy (emphasis mine) Not "against the letter of policies and guidelines". "Against the letter of policy". (Of course, you could also be referring to WP:EPAI, which is already policy that bans most uses of LLMs.) SuperPianoMan9167 (talk) 21:08, 5 November 2026 (UTC)
I was not using the word "policy" precisely there but as a shorthand for "policies and guidelines", which given the context of this discussion was rather silly of me. Thryduulf (talk) 21:29, 5 November 2026 (UTC)
Understood. SuperPianoMan9167 (talk) 21:30, 5 November 2026 (UTC)
AI uses WP to train itself No. Humans train AI using WP. Or direct AI to train itself using WP. Either way, humans are the ones deciding to use WP as training material. SuperPianoMan9167 (talk) 21:04, 5 November 2026 (UTC)
I think we could cover both sides of the coin, "Wikipedia isn't a slop farm" and "Wikipedia isn't a Butlerian jihadist sleeper cell" Kowal2701 (talk, contribs) 18:37, 5 November 2026 (UTC)
I have been thinking about "by humans for humans" and I can't support it. Discussion has focused on "by humans" but I think that's acceptable. We are comfortable saying "Wikipedia is not a business directory" even though, gods know, parts of it are. But "for humans" contradicts our license, which clearly permits use of Wikipedia as training data; it also contradicts the considered decision not to (try to) block scrapers. This is too far from the truth for a slogan, as far as I'm concerned.
I also think that if we promote NOLLM to policy, we should explicitly address use of LLM to locate sources. I have seen Teahouse hosts flatly say this is not allowed, which I believe is unjustified. "Use of LLMs to search for sources is allowed, but editors are responsible for reading, evaluating, and summarizing sources themselves." might stop a lot of arguments in their tracks. M kuhner (talk) 21:58, 10 November 2026 (UTC)
I can't see how "for humans" purports to limit other uses. It's a statement of our target audience: Wikipedia articles are written with their primary goal to be read by people. If I may wax snarky, contrast Wikidata (by humans, for Google Knowledge Graph; but clearly not human-targeted in any case).
It would be a stretch to call LLM use solely to locate sources "generat[ing] or rewrit[ing] article content". But even if you read "generate" expansively enough to include research with no text carried forward, then it wouldn't be detectable anyway, so it's moot. And no, that's not an argument to scrap or sideline NOLLM, any more than the fact that it's so difficult to detect copyright infringement via translation is reason to rescind copyright policy. —Cryptic 23:53, 10 November 2026 (UTC)
I wouldn't fuss over the guideline (lack of guideline) on the use of LLM for source-finding if it were not for multiple instances where it has not seemed to be clear enough. The infeasibility of preventing it is quite often used as an argument against NOLLM as a whole; and people are misinformed at the Teahouse about allowability of LLM searches on a somewhat regular basis. Let's make the guideline/policy clear on this point. I don't think it will take very many words. M kuhner (talk) 00:06, 11 November 2026 (UTC)
I've seen instances of people (presumably) reading "NOLLM" and thinking that means no using LLMs at all in any way for anything related to Wikipedia. There are some who have made it clear that they would like that (or something closely approximating that) to be the rule, but it is not the rule as things stand. Thryduulf (talk) 00:22, 11 November 2026 (UTC)
WP:LLMRESP did used to be pointed to at the start, now it's a hatnote, I've tried to make it clearer in Special:Diff/1384147012/1384955946 Kowal2701 (talk, contribs) 01:06, 11 November 2026 (UTC)

RFC on new notability proposal adoption

 – 🪐Kepler-1229b | talk | contribs🪐 20:35, 7 November 2026 (UTC)

Should this proposal be adopted as an official notability criterion? Should the proposed notability criterion WP:NLANG be adopted?

A: Yes
B: No

Changes in wording or substance may also be discussed here; if consensus is reached for certain changes they will also be implemented. Pinging previous participants in discussions about this: @Kwamikagami: @Boynamedsue: @Sirfurboy: @TaivoLinguist: @Cnilep: 🪐Kepler-1229b | talk | contribs🪐 04:45, 17 October 2026 (UTC)

In addition, new proposals for such a notability criterion can also be made here; if they gain consensus as well they may also be adopted. 🪐Kepler-1229b | talk | contribs🪐 01:41, 12 November 2026 (UTC)

Survey

Aren't we supposed to be answering A (yes) or B (no)?

  • B I don't like that this depends so much on the idea of a language being stable when my understanding is that languages are not stable over time. And the distinction between a language and a dialect seems arbitrary as they are a continuum. Andrew🐉(talk) 20:31, 11 November 2026 (UTC)
    The language only needs to have been transmitted between generations for any period of time to qualify here. 🪐Kepler-1229b | talk | contribs🪐 20:36, 11 November 2026 (UTC)
  • B: No we do not need an SNG here, if a language is notable it will surely pass the GNG threshold. Cavalryman (talk) 06:41, 12 November 2026 (UTC).
  • B: No We don't need another obscure notability guideline when the GNG is sufficient. Wikipedia already has too many special rules. Joe vom Titan (talk) 18:57, 12 November 2026 (UTC)

Discussion

I wonder if the stability section should come before confirmation, given that it is so much shorter. The layout would be easier to follow that way.
Maybe a few wording tweaks for clarification (e.g. intermediate nodes in a classification), but I don't see any substantial problems. — kwami (talk) 05:22, 17 October 2026 (UTC)
I don't see what else we need to add there to be honest. Got any suggestions? 🪐Kepler-1229b | talk | contribs🪐 05:25, 17 October 2026 (UTC)
Yeah, actually, I guess not. I tried reversing the word order of proposed-but-not-generally-accepted language families and intermediate nodes, which if we're being pedantic should be hyphenated, but it wasn't an improvement. — kwami (talk) 05:37, 17 October 2026 (UTC)
Makes sense. Also can't see any more problems here. 🪐Kepler-1229b | talk | contribs🪐 05:40, 17 October 2026 (UTC)
I'm still a little concerned about the article's acceptance of Glottolog alone. I would also be a little concerned about dialects not being notable as a principle.Boynamedsue (talk) 06:22, 17 October 2026 (UTC)
How is it Glotto alone?
If dialects are inherently notable, how do we identify what counts as a dialect? It's an open-ended category with AFAICT no criteria for inclusion. Basically anything could be claimed to be a "dialect" and therefore warrant an article. — kwami (talk) 08:53, 17 October 2026 (UTC)
My interpretation of the proposed guideline is that a Glottolog entry is enough for notability, unless reliable sources state that the language has been given an entry in error. Ironically, in my view, sources discussing why a language doesn't exist make the purported language more notable than a language only supported by a Glottolog page.Boynamedsue (talk) 11:46, 17 October 2026 (UTC)
That's the crux of the issue. IMO a people, a nation, a language are all inherently encyclopedic. A spurious language may have more coverage than an obscure real one, but IMO we shouldn't dismiss entire peoples as not worthy of coverage. A spurious language can be included too if the sources support it.
It's not that Glotto is in itself adequate for notability, but that they demonstrate existence, and it is the language itself that establishes notability. — kwami (talk) 19:08, 17 October 2026 (UTC)
I don't really believe in peoples and nations. But I would agree that a language is almost always notable. I just don't believe that we can be sure a language exists from the fact it has a glottolog entry.--Boynamedsue (talk) 21:22, 17 October 2026 (UTC)
@Boynamedsue They do have verification criteria, which you can see here. 🪐Kepler-1229b | talk | contribs🪐 21:29, 17 October 2026 (UTC)
From what I've seen in the past, Glottolog has written entries based on fragmentary centuries old wordlists. For me, it's not sufficient on its own.--Boynamedsue (talk) 21:48, 17 October 2026 (UTC)
Only if there's enough data to conclude that it was a distinct language. — kwami (talk) 21:50, 17 October 2026 (UTC)
In my view they have concluded a language exists in situations where there isn't enough data to do so, like that of Magiana language. As discussed above, they have likely made a mistake by classifying Magiana separately from Majena language. I personally don't feel comfortable using Glottolog alone for poorly attested languages, as they sometimes base their entries solely on very old primary sources that have not been analysed by relevant linguists. I won't be able to support this proposal unless Glottolog is removed.--Boynamedsue (talk) 03:44, 18 October 2026 (UTC)
@Boynamedsue They haven't because they don't list both Magiana and Majena. They only list one of the two. 🪐Kepler-1229b | talk | contribs🪐 04:05, 18 October 2026 (UTC)
Yes, that is true. It doesn't seem to be aware of Majena, which is better covered by scholars. In my view this doesn't massively change my point; Glottolog is a reliable source, but not a good enough one to have a special exception from WP:GNG carved out for it.Boynamedsue (talk) 04:18, 18 October 2026 (UTC)
You may wish to have a look at the proposal further down if you have not done so. 🪐Kepler-1229b | talk | contribs🪐 05:34, 18 October 2026 (UTC)
I don't think the RfC is particularly well formed. What is "this proposal"? Editors who have not participated in the preceding discussion will appreciate a clearer statement of the question. Regulov (talk) 15:39, 17 October 2026 (UTC)
The proposal is the project page for which this is the discussion page, but I agree that this is unclear. I have had to double check this twice. Is this the correct venue for such a significant RfC?Boynamedsue (talk) 15:48, 17 October 2026 (UTC)
Since there haven't been any votes yet, perhaps the RfC could be held off until it is clarified? Kisseran (talk) 16:13, 17 October 2026 (UTC)
Apologies for the poorly worded RfC; this is my first time opening one and I don't ave any experience with them. 🪐Kepler-1229b | talk | contribs🪐 18:59, 17 October 2026 (UTC)
+1 a link to RFCBEFORE would be helpful. Dw31415 (talk) 21:38, 7 November 2026 (UTC)
Previously, I had discussed with other editors on the proposal before beginning the RfC. 🪐Kepler-1229b | talk | contribs🪐 21:44, 7 November 2026 (UTC)

I read the proposal on the page, but I find it unclear. It says languages should be "stable", but it does not tell us how we are supposed to determine that easily; then it talks about "confirmation" but of what? I think WP:GNG is more clear and sufficient. --Piotr Konieczny aka Prokonsul Piotrus| reply here 00:49, 18 October 2026 (UTC)

Both of those are explained.
"Stable" means transmitted across generations. Confirmation is confirmation that the language exists. — kwami (talk) 00:56, 18 October 2026 (UTC)
Which can and should be determined through RS, and there is no need for anything but GNG IMHO. Can you show me an example of an article that should be on Wikipedia, common-sense wise, but does not pass GNG meriting this proposal? Piotr Konieczny aka Prokonsul Piotrus| reply here 01:05, 18 October 2026 (UTC)
It's not clear to me that this is useful. In the past 12 months there have been 12 AfDs on languages or language families, and 5 of those (language groupings and conlangs) would be covered by the GNG and not this proposal. 2 of the 12 were scope issues that this proposal probably wouldn't have helped with. The remaining 5 cases don't clearly align with the proposal's likely outcomes for entries with minimal coverage beyond a database entry. Of those 5, 2 had glottolog codes (a third had a LinguistList code) but were merged or renamed because the language itself didn't have significant coverage. The proposal would have been clearly useful 5 times in the last year, and might have gotten the opposite result in 2 or 3 of those 5 cases.
Lack of significant coverage
Wikipedia:Articles for deletion/Shoccoree-Eno language merge no ISO or glottolog code, did have a LinguistList code.
Wikipedia:Articles for deletion/Mbwasa language redirect no database codes
Wikipedia:Articles for deletion/Magiana language merge glottolog code
Wikipedia:Articles for deletion/Romani Tatar language delete no database codes
Wikipedia:Articles for deletion/Noric language renamed glottolog code, significant coverage of the inscriptions, but not for the language
Conlangs which fail the GNG
Wikipedia:Articles for deletion/Bolak language delete
Wikipedia:Articles for deletion/Moss (language) redirect
Proposed language families which fail the GNG
Wikipedia:Articles for deletion/Paraguay–Uruguay Sign Language family delete created only on the basis of a glottolog code, deleted after glottolog update
Wikipedia:Articles for deletion/Chukotko-Kamchatkan–Amuric languages redirect
Wikipedia:Articles for deletion/Uralo-Siberian languages merge
Scope issues
Wikipedia:Articles for deletion/Hellenic languages merge due to overlap with Greek language. Both pages listed the same database codes.
Wikipedia:Articles_for_deletion/Southern_Kayapó_language merge as a non-notable language era. No independent database codes.
Altogether, I don't really see why this is needed, and even if it is, I don't think it's workable in its current state. Others have pointed out conflicts with parts of the notability policy so I won't repeat those. — Wug·a·po·des 01:21, 18 October 2026 (UTC)
Just to add to your analysis, I think you have missed at least six AfD discussions in the past 12 months because they do not include the word "language" in their titles:
Wikipedia:Articles for deletion/Vozgian
Wikipedia:Articles for deletion/Adjuvilo
Wikipedia:Articles for deletion/Humanist Greek
Wikipedia:Articles for deletion/Modern Meitei
Wikipedia:Articles for deletion/Ningombam Bupenda Meitei
Wikipedia:Articles for deletion/Andro (Meitei dialect) Kisseran (talk) 01:49, 18 October 2026 (UTC)
I think there's some more Meitei dialect stuff that's been missed so far. 🪐Kepler-1229b | talk | contribs🪐 01:50, 18 October 2026 (UTC)
Could someone give me a tl;dr on the background of this proposal? If I understand correctly, this proposal seeks to do two things:
  1. set a somewhat lower bar for inclusion for languages by auto-keeping language articles that satisfy this proposal even if they fail GNG.
  2. clarify our stance on what a listing in reference works like Glottolog and Ethnologue contributes to notability for languages/language varieties.
Is that an accurate understanding of? Since the intent seems to be to tilt the scales towards inclusion, could the proposer(s) link my to a couple AFDs that closed delete that should have closed keep in the view of the proposer(s)? -- LWG talk (VOPOV) 11:58, 13 November 2026 (UTC)
@LWG Draft:Baazəm language for sure, and possibly Magiana language and Shoccoree-Eno language as well, though those two are edge cases. 🪐Kepler-1229b | talk | contribs🪐 13:43, 13 November 2026 (UTC)
Tour understanding seems to be more or less correct as well. 🪐Kepler-1229b | talk | contribs🪐 13:43, 13 November 2026 (UTC)
This was intended as something we could point to when someone mass-tags language articles for deletion. There weren't problems with articles actually being deleted at AFD, because the argument that languages are notable had been enough. Putting it in writing just saved the effort of going through the same thing over and over. But now we've had a successful AFD proposal for Baazəm, which goes against 20 years of precedent that languages are inherently notable. — kwami (talk) 19:25, 13 November 2026 (UTC)

Alternative proposal to the same effect

I think the current proposed guideline based on the confirmed existence of the language alone is unlikely to be convincing enough to pass. Instead, I think the guideline should take GNG as a starting point, and explicitly prescribe the amount of known information that would be sufficient to justify a standalone article, as opposed to being a section in an ethnographic article or an entry in a list. Here's an alternative proposal, which I think will have largely the same effect, but easier to justify:

A human language is presumed notable if it satisfies GNG or three or more of the following are explicitly covered across all reliable sources:

  1. A Glottocode or ISO-639 code
  2. A numerical estimate for speaker count (at any point in time) that is nonzero
  3. A phonological or phonetic inventory
  4. Any morphological feature or combination thereof, that if presented in tabularly, contains at least three distinct items
  5. Any description of word order
  6. A word list of at least two words, accompanied with translations into the language of the relevant source
  7. An acknowledgement of the absence of information for any of the above in a source that supplies information for any other one of the above

For the purposes of notability, Glottolog and any source listed therein is presumed to be independent and reliable, and significant coverage as outlined above is sufficient even if only a single reliable source exists.

[Commentary: These criteria obviate the problem of presuming notability based on existence and the concern of worthless permastubs. The exact wording and requirements are open to further refinement, of course. These can also be applied to dialects, reconstructions, and constructed languages naturally with the provisions proposed below]

A dialect, sociolect, or any other language variety is subject to the same criteria as above, with the provision that the relevant passage in a source must explicitly refer to the variety in question. A work that purports to be description of an entire language X does not provide notability for a variety Y if the work exclusively treats Y.

A constructed language is subject to the same criteria as above, with the provision that the relevant reliable sources must be independent from the creator(s) of the language. A source written in a constructed language is presumed to be non-independent from that constructed language.

A reconstructed language is subject to the same criteria as above, with the provision that a scholar's own works are not independent from their own reconstruction.

A grouping of languages is presumed notable if its reconstructed protolanguage is notable, or if a single work treating the group as a whole is significant enough to meet the criteria above for two or more members of the group in question.

Kisseran (talk) 14:41, 17 October 2026 (UTC)

"A work that purports to be description of an entire language X does not provide notability for a variety Y if the work exclusively treats Y."
Are you sure this is strictly what you intend? I am not sure I follow the sense. Does "exclusively" mean the work does not discuss other varieties? Does it mean the work fails to describe the entire language X of which Y is understood to be a variety? Why would such a failure mean the discussion of Y does not count toward establishing Y's notability? I don't get it. Regulov (talk) 15:29, 17 October 2026 (UTC)
Yes, this is what I intend. Due to practical limitations, many descriptive grammars of some language are based mostly or exclusively on a single variety of that language, while ignoring other varieties. For instance, the 2015 Grammar of the Yakkha Language begins its section on dialectal variation thus:

The variety documented here is spoken in Tumok village and surrounding areas, e.g., in Salle. No detailed dialectal study has been undertaken for Yakkha yet. Based on phonological differences and distinct exclamative words, I tentatively propose three further dialects: one spoken in the area around Ankhinbhuin (Angbura, Hombong, Phakling), one spoken in the area around Dandagaun and one spoken towards the north, in Kingring and Kharang villages.

Now, the grammar does not go on to provide enough detail to establish notability for these additional dialects by the standards I have just proposed. This is a good, intended outcome, since the information is sparse enough that it would fit better as a section in the article Yakkha language. However, one may argue that since the rest of the 600+ page grammar treats the Tumok variety of Yakkha, that an article like Tumok Yakkha is warranted. In the absence of other sources, this would be a bad idea, since such an article would almost entirely duplicate the current article on the Yakkha language. Kisseran (talk) 16:04, 17 October 2026 (UTC)
I think I understand now. Thank you. Consider rephrasing. I take it the important thing is that a variety or dialect's notability (distinct from that of its parent) requires at least one other notable, clearly distinct sibling variety or dialect. In a discussion of language A, dialects P(A) and Q(A) are covered. On this basis, the discussion counts toward the notability of both dialects. In another discussion of language A, only one dialect, P(A), is mentioned; and in a third, only Q(A). These latter discussions, under the rule, cannot be used to demonstrate the notability of either dialect.
"A work purporting to describe an entire language X does not provide notability for a variety Y if Y is the only variety of language X it covers."
"In order for a work purporting to describe an entire language X to provide notability to a variety Y, it must cover at least one other variety of the same language in comparable detail."
"A work that purports to be description of an entire language X does not provide notability for a variety Y if the work does not cover at least one other variety of X in similar detail."
Regulov (talk) 17:07, 17 October 2026 (UTC)
According to one reading of these criteria, the "language" of nearly every city or conurbation in the world is worthy of an article. For Southern California, we could have articles on San Diego English, El Cajon English, Los Angeles English, Riverside English, Santa Barbara English, and likely dozens more. None of these are actually notable, IMO, but they could easily fit the criteria.
Actually, we might have the same for an ever-expanding number of schools and colleges.
On the other hand, if we adhere strictly to the criteria above, English fails, because Glottolog is a RS but doesn't cover at least three of the above. All languages fail these criteria, and we'd need to fall back to GNG and there is therefore no point to this proposal.
The guideline as currently worded reflects the actual criteria as currently practiced on WP. Its purpose is merely to have a remedy against new editors tagging hundreds of languages as non-notable, and demanding to know why their tags were mass-reverted. It doesn't need to be an official guideline to accomplish that. If the choice is between the current state of affairs and corrupting it to uselessness because we anticipate that there will be no end of contradictory objections, then I vote to leave things as they are. They're working just fine, and I don't see any benefit to changing them. — kwami (talk) 19:29, 17 October 2026 (UTC)
I would be curious to see the sources that you would point to to show that Riverside Californian English is notable under this proposal. On the other hand, I'm not sure how you take the criteria to exclude English. Are you interpreting "at least three or more across all reliable sources" as "at least three of more in every reliable source"?
The current drafted guideline will not pass. It flies in the face of WP:NRV which states No subject is automatically or inherently notable merely because it exists. WP:LOCALCONSENSUS states that unless they can convince the broader community that such action is right, participants in a WikiProject cannot decide that some generally accepted policy or guideline does not apply to articles within its scope. If you want to mass-revert notability tags, you need to back that up with a relevant policy or guideline. Kisseran (talk) 20:01, 17 October 2026 (UTC)
I feel like the points for 🪐Kepler-1229b | talk | contribs🪐 20:19, 17 October 2026 (UTC)
A phonological or phonetic inventory
Any morphological feature or combination thereof, that if presented in tabularly, contains at least three distinct items
Any description of word order
A word list of at least two words, accompanied with translations into the language of the relevant source
could be condensed into a general description of the language in question. 🪐Kepler-1229b | talk | contribs🪐 20:20, 17 October 2026 (UTC)
Also, the two-word limit seems a bit arbitrary when there are languages such as Arma–Pozo language which are known from exactly a single word (the example in question has been covered in sufficient sources to warrrant an article). 🪐Kepler-1229b | talk | contribs🪐 20:23, 17 October 2026 (UTC)
Additionally, "A numerical estimate for speaker count (at any point in time) that is nonzero" seems a bit dense and unwieldy for a notability description; it could be condensed into "The language has been spoken by any number of people at any point of time" or something like that. 🪐Kepler-1229b | talk | contribs🪐 20:35, 17 October 2026 (UTC)
Replying point by point:
  • I divided "a general description of the language" into multiple points because people will disagree with respect to what counts as a "general description", and it wouldn't be much of an improvement compared to GNG's "significant coverage". Consider that you will be applying this guideline to convince people at AfD, most of whom will not have much linguistic expertise.
  • I'm not super attached to the two-word limit. If people think "at least one word is known" would fit better, then that would be fine as well. That said, I don't think Arma-Pozo language is a particularly compelling example, considering that its ISO-639 code was deprecated because there wasn't enough evidence for its existence. It would do fine regardless, since it meets GNG, but at that point the article is more about a historical anecdote than linguistic scholarship.
  • "A numerical estimate for speaker count (at any point in time) that is nonzero" is worded like that because simply knowing that some number of people spoke it at some point in time is tantamount to knowing that it exists, whereas knowing approximately how many people spoke it is something that you should be able to point to as significant coverage.
Kisseran (talk) 21:06, 17 October 2026 (UTC)
I can see where you are coming from here, and I agree with your points as worded currently. Nevertheless, I do feel like points for the documentation should be grouped somehow together, perhaps indented together somehow. 🪐Kepler-1229b | talk | contribs🪐 21:13, 17 October 2026 (UTC)
And again, how do we avoid non-notable "dialects"? I'm thinking of language activists (India and Pakistan come to mind, but there are others) who insist that their "language" warrants an article, even though it is not established as a language or even a distinct dialect. If I can show an inventory (exactly the same as the parent language/dialect), a couple regionalisms, or even just words that duplicate the parent language/dialect, and the population of the district, then my "language" meets the notability criteria. — kwami (talk) 20:35, 17 October 2026 (UTC)
I do find myself partial to the proposal of at least two dialects meeting notability to have articles about them. 🪐Kepler-1229b | talk | contribs🪐 20:49, 17 October 2026 (UTC)
Yes, I do like that. But do we have any dialect articles where it's the same dialect as the language article because nothing else is attested? I suppose we might have cases where a dialect is reported to be divergent, and is the only one that is attested well enough for an article. But then it probably shouldn't be merged with the language article as if it were representative. — kwami (talk) 21:48, 17 October 2026 (UTC)
Not dialects, but we do have comparable recent precedent in Wikipedia:Articles for deletion/Hellenic languages and Wikipedia:Articles for deletion/Modern Meitei. Kisseran (talk) 21:54, 17 October 2026 (UTC)
If people insist that a language variety warrants an article, and they can find reliable sources to support it, then it's no longer a problem of notability. There are other policies and guidelines, such as WP:POVFORK and WP:NOTSOAPBOX that circumscribe the range of appropriate article beyond notability. Also, I can't think of an instance of such a spurious, non-notable dialect that would be allowed by the proposed criteria, but fail GNG. Even if someone does manage to publish such a spurious description, it is entirely feasible to argue that it would be unreliable for the reasons that you suggest. Kisseran (talk) 20:50, 17 October 2026 (UTC)
This would certainly be more clear. Do ping me if this is formally suggested for a vote. Piotr Konieczny aka Prokonsul Piotrus| reply here 00:50, 18 October 2026 (UTC)
Couldn't support this. We correctly !voted to merge Wikipedia:Articles for deletion/Magiana language, which is incorrectly listed with a Glottolog code despite being a synonym of a better known language that is not listed, Majena language. We would have been constrained to accept it on the basis of 2 and 6 appearing in a primary source from the 1790s if this proposal was accepted. I would want to remove Glottolog from the list and add the qualifier "in a reliable secondary source or the research of a scholar with academic credentials." to numbers 2-7.Boynamedsue (talk) 07:05, 18 October 2026 (UTC)
Does 2 appear in the source? I'm not sure how to gain access to it but I don't see any mention of speaker count in the merged article. Kisseran (talk) 11:04, 18 October 2026 (UTC)
No figure for speaker count seems to have been ever given for Magiana/Majena. 🪐Kepler-1229b | talk | contribs🪐 17:54, 18 October 2026 (UTC)
I thought there was a number of villages and people given in one of the 18th century sources? Must be misremembering. In any case, in my view, as we have seen errors with Glottocodes, I reject giving it any special privileges.Boynamedsue (talk) 05:08, 19 October 2026 (UTC)
Glottolog errors can be easily corrected by posting to their Github page, so I don't see the problem here. I've done that quite a lot and they have added the correct changes. I further quote from Majena: There is some confusion between the Majena-speaking Ticomeri and another group, also known as "Ticomeri", who spoke a divergent dialect of Moxo. Whether the two groups were related (i.e. whether the Ticomeri had abandoned Majena and acquired Moxo) is unknowable, since both were apparently extinct by 1805 (emphasis mine). 🪐Kepler-1229b | talk | contribs🪐 06:12, 19 October 2026 (UTC)
Well in Wikipedia we work with reliable sources, not just perfect sources. In any case, the heuristic proposed here makes the correct determination, and I find it highly unlikely that any erroneously assigned Glottocodes will be referenced to enough information to meet two of the other six criteria. Kisseran (talk) 08:29, 19 October 2026 (UTC)
Even notable subjects can be merged to other articles per WP:PAGEDECIDE. Katzrockso (talk) 05:21, 11 November 2026 (UTC)

A thought

I am not very familiar with languages on Wikipedia, but I am somewhat familiar with GNG-issues. While I don't necessarily agree that language should be auto-notable, I am also not sure how their case differs, common sense, from some topics we have exceptions for, ex. taxa (species, etc.), astronomical objects (or the perennial issue of minor sport people and events). It might be worth looking into how their relevant GNG exceptions are worded. --Piotr Konieczny aka Prokonsul Piotrus| reply here 01:34, 18 October 2026 (UTC)

Completely agree, as I also do work with astronomical objects sometimes as well. 🪐Kepler-1229b | talk | contribs🪐 01:36, 18 October 2026 (UTC)
One very good reason why we would treat them differently is that there is always a debate as to what constitutes a language and what constitutes a dialect in any dialect continuum. Although I think the vast majority of languages will be notable, I would say that these are going to be clearly notable because people have written about them. GNG is sufficient. A notability guideline can make a useful shortcut towards a presumption of notability, but in the case of languages, speakers of Atwillery may differ with speakers of Atwollery as to whether the latter constitutes a language. As Atwollery speakers have no army, they cannot press their case by conventional means. They thus turn to Wikipedia, and if we have this guideline, they can create the Atwollery page, even though no one in the world has written about their "language" (Well, except for the Atwollery Language League).
Of course, the guideline proposes that languages are notable and dialects are not. How do we know which Atwollery is? Only with reliable secondary sources that make the case. Without such, I don't think this guideline settles anything. With such, we can make the case on GNG alone. Sirfurboy🏄 (talk) 12:32, 18 October 2026 (UTC)
But that's exactly the role that the word "confirmed" plays: lects are not languages just because their speakers say so, but because 2ary RS's confirm them as languages. — kwami (talk) 21:43, 7 November 2026 (UTC)
@Blueboar I can list Draft:Baazəm language, which is evidently used as the primary language of a speech community, but not properly discussed in any reliable source due to insufficient documentation. 🪐Kepler-1229b | talk | contribs🪐 22:12, 7 November 2026 (UTC)

Jew Tagging

What (and more to the point, where) is the Wikipedia policy against Jew tagging—the practice of identifying as Jewish every named individual to whom Jewishness can be attributed, often as defined to the Nuremberg Laws? For example, should all sportscasters or all baseball players who have at least one Jewish grandparent be identified as Jewish? MarkBernstein (talk) 22:52, 12 November 2026 (UTC)

@MarkBernstein, there's an essay: Wikipedia:Jew-tagging. It includes a helpful table that links to discussions on the subject. Schazjmd (talk) 22:59, 12 November 2026 (UTC)
I don't think there's a policy on this per se but people have been blocked for doing this in the past, see here. ♠JCW555 (talk)♠ 23:02, 12 November 2026 (UTC)
Not a policy, but there's a relevant guideline, WP:CATREL, which among other points (demanding a personal statement of faith) says "Never categorize by a religion of any parents or other ancestors." —David Eppstein (talk) 01:13, 13 November 2026 (UTC)
There was also a long discussion about this last year, with further discussion 9 months ago. I see you participated in the latter. Anomie⚔ 01:59, 13 November 2026 (UTC)
The policy is WP:NPOV, and in particular WP:DUE. We don't need a specific 'don't engage in Jewtagging' policy, since the community has already made it clear enough in the recent discussions linked above that such behaviour is inappropriate, distorting, and contrary to the purpose of an encyclopaedia that aims to reflect proportionally coverage in relevant reliable sources that directly relates to notability etc. Trying to frame a specific policy would prove problematic anyway, given that if, when, where, and how it is appropriate to note ethnicity and/or faith may sometimes come down to editorial judgement. Anyone insisting that an article must assert that a person is Jewish merely because that person has a Jewish grandparent is already violating Wikipedia policy: we don't need a new one. Anyone slapping the bald statement 'X is Jewish' (citing a source) into a biography is either violating WP:DUE, or needs to learn how to write if there is actually a legitimate reason to discuss the matter. Beyond that, context matters. We shouldn't need to inform our readers that Rabbi Ishmael was Jewish, and indeed we don't, thankfully (personally, I'm opposed to treating our readers like idiots), but clearly there are times when it is relevant and needs to be stated explicitly. If so, the article can explain why. If it can't, it's Jew-tagging. If relevance is in question, it can be discussed - with the onus to provide evidence falling on those wishing to include such statements, and with sources merely engaging in Jew-tagging themselves (for whatever reason) given little or no consideration. WP:DUE, once again - content needs to be justified, and not just cited, regardless of what it concerns, and Wikipedia can de better than the one-dimensional labelling that Jew-tagging entails. AndyTheGrump (talk) 03:03, 13 November 2026 (UTC)

Loosen WP:BIDI pertaining to sidebars

WP:BIDI is part of the Wikipedia:Categories, lists, and navigation templates editing guideline. I propose the following revision:

−
Every article that [[Help:Transclusion|transcludes]] a given navigation template should normally also be included as a link in the template, and every navigation template that includes a link to a given article should normally also be transcluded onto that article, so that navigation is bidirectional.
+
Every article that [[Help:Transclusion|transcludes]] a given navigation template should normally also be included as a link in the template, and every navigation template that includes a link to a given article should normally also be transcluded onto that article, so that navigation is bidirectional. Editors may use additional discretion when deciding whether a sidebar navigation template must be bidirectional. For example, if introducing a sidebar to an article would be more disruptive than helpful (such as by introducing significant visual clutter to the page), editors may include a unidirectional link from the sidebar to the article.

Recent history of WP:BIDI discourse:

The suggested proposal would correct these issues by loosening WP:BIDI as it pertains to sidebars.

~ Alexandraaaacs1989 (talk) 18:31, 18 November 2026 (UTC)

Oppose as strongly as possible. If anything, we need to strengthen it and remove the ambiguity. We are already plagued by a number of tangential, badly constructed sidebars that take up massive amounts of space while adding no useful encyclopedic content. If it clutters the page or frustrates editors, then why do we need it in the sidebar? PARAKANYAA (talk) 18:50, 18 November 2026 (UTC)
The idea is that loosening the guideline would encourage editors to focus less on WP:BUREAUCRACY and more on evaluating content on the basis of its merits ~ Alexandraaaacs1989 (talk) 19:17, 18 November 2026 (UTC)
The merits of a sidebar is that it is a tightly related collection of links. None of this is that. PARAKANYAA (talk) 19:40, 18 November 2026 (UTC)
I think a bigger problem is that people very often disagree on what constitutes "tightly related". SuperPianoMan9167 (talk) 19:41, 18 November 2026 (UTC)
Yes: a "tightly related" group of 500 links, because 500 articles could benefit from the sidebar's links to a handful of key articles, but BIDI is being enforced to require that all 500 displaying articles also be listed in the sidebar. WhatamIdoing (talk) 22:46, 18 November 2026 (UTC)
That's what "See also" is for. Those 500 articles are not "part of a series". Sirfurboy🏄 (talk) 22:57, 18 November 2026 (UTC)
Here's the problem I see:
  • Alice: Oh, this article is about ____. People reading this article may be interested in a quick way to get to the main articles about ____. I'll add the sidebar.
  • Bob: WP:BIDI says you can't add the sidebar about ____ unless you put [this article] in the sidebar. I'm adding [This article] to the ____ sidebar. Now there are 11 articles in the ____ sidebar.
  • Chris: People reading this other article may be interested in a quick way to get to the main articles about ____. I'll add the sidebar.
  • Bob: WP:BIDI says you can't add the sidebar about ____ unless you put [this other article] in the sidebar. I'm adding [This other article] to the ____ sidebar. Now there are 12 articles in the ____ sidebar.
  • David: People reading this article over here may be interested in a quick way to get to the main articles about ____. I'll add the sidebar.
  • Bob: WP:BIDI says you can't add the sidebar about ____ unless you put [this article over here] in the sidebar. I'm adding [This article over here] to the ____ sidebar. Now there are 13 articles in the ____ sidebar.
  • ...
  • Yael: People reading this little article may be interested in a quick way to get to the main articles about ____. I'll add the sidebar.
  • Bob: WP:BIDI says you can't add the sidebar about ____ unless you put [this little article] in the sidebar. I'm adding [This little article] to the ____ sidebar. Now there are 35 articles in the ____ sidebar.
  • Zed: People reading this new article may be interested in a quick way to get to the main articles about ____. I'll add the sidebar.
  • Bob: WP:BIDI says you can't add the sidebar about ____ unless you put [this new article] in the sidebar. I'm adding [This new article] to the ____ sidebar. Now there are 36 articles in the ____ sidebar.
...at which point, we have not just clutter, but clutter caused by thoughtless enforcement of WP:BIDI, when what was actually wanted in all of these cases was just links to the main articles about ____ and not links to all the non-main articles that could benefit from a link to just the main articles.
I support the proposed addition, and I suggest additionally the following two principles:
  1. Sidebars should be short. I don't know if we should define 'short' as a maximum of 20 links or 50 links, but it should not be infinitely expandable, and it probably needs a specific number stated as an anchor. Sidebars that currently contain hundreds of links should either be converted to footer navboxes, or split to create a shorter sidebar with the main articles and a longer (BIDI-compliant) footer navbox with all the articles.
  2. Sidebars should focus on major articles. For example, if the subject of the sidebar is Philosophy then the sidebar should have links History of philosophy, Epistemology, Ethics, Logic, Metaphysics but links to sub-subtopics such as Type physicalism, Mind–body problem, and Neutral monism should be relegated to footer navboxes. In some cases, the sidebars will be BIDI-compliant or nearly so (e.g., if the subject of the sidebar is a particular book series, then the only links wanted in the template are author+the books, and the sidebar could reasonably be used only on those articles), but this will be neither required nor expected.
WhatamIdoing (talk) 19:31, 18 November 2026 (UTC)
I can understand the problem you're looking to solve, and it's somewhat different (and opposed) to the problem OP is looking to solve. I think that to satisfy your concern it would be sufficient to loosen BIDI to allow transclusion of the template without including a link to the page on which it's transcluded.
I think OP's concern is primarily with the reverse case, that transclusion of the template might be opposed on grounds unrelated to the template itself but based solely on considerations relating to the page (e.g. that it's already cluttered with other sidebars). This would mean that the link would have to be excluded despite its potential utility to the template.
I actually agree with both concerns, but they are opposed in that your concern is to allow the utility of templates on pages they don't link to while keeping their size manageable, and OP's would allow for the unchecked expansion of templates.
I'm not opposed to loosening BIDI to accomodate both, but I think a safeguard like the numerical limit would be vital, as well as to make template guidelines stricter in CTOPs. That could allow us to add the flexibility for templates to work well in 90% of the project while preventing the misuse of that flexibility in the 10% that's subject to the most intense POV-pushing. Samuelshraga (talk) 20:00, 18 November 2026 (UTC)
I think OP's concern is primarily with the reverse case, that transclusion of the template might be opposed on grounds unrelated to the template itself but based solely on considerations relating to the page (e.g. that it's already cluttered with other sidebars). This would mean that the link would have to be excluded despite its potential utility to the template. Exactly yes, this is my primary concern
OP's [suggestion] would allow for the unchecked expansion of templates An important point I think is relevant here is that unchecked expansion of sidebars is currently permitted, and this guideline would maybe slightly exacerbate the issue, but it by no means would be a dealbreaker. I agree that clearer guidelines about what constitutes a tightly-related article seem due considering the amount of disagreements that seem to arise over tight relatedness, and I think this would be a useful idea to examine in subsequent VPPs (though I find a hard numerical ceiling to be a bit too harsh) ~ Alexandraaaacs1989 (talk) 20:34, 18 November 2026 (UTC)
I think that the "unchecked expansion of sidebars" is something we need to stop. WhatamIdoing (talk) 22:48, 18 November 2026 (UTC)
I reverted the April 2026 bold change to the wording of WP:BIDI that encourages more "unchecked expansion of sidebars". As I noted in the April 2026 RfC , past discussions have not reached any conclusion about altering the wording of WP:BIDI to make it truly bidirectional (i.e. requiring any page with a navbox to be an entry in that navbox). Katzrockso (talk) 22:58, 18 November 2026 (UTC)
I don't think we need "a hard numerical ceiling". I do, however, think that we need a number. "About 20 links" is not "a hard numerical ceiling", but it should give an editor a clear indication that if they're putting 100 links in the sidebar, they're screwing up (and need to switch to the footer style). WhatamIdoing (talk) 23:30, 18 November 2026 (UTC)
@Samuelshraga, thanks for that note. You're correct that I don't really want a "unidirectional link from the sidebar to the article". Instead, I want to permit the inclusion of the sidebar in a relevant but peripheral article without bloating the sidebar to include a link to the peripheral article. It should be unidirectional out of the peripheral articles, rather than unidirectional into 100+ peripheral articles. WhatamIdoing (talk) 23:38, 18 November 2026 (UTC)
The current proposal is to allow unidirectional links from the sidebar to the article, so this would mean you oppose the proposal. SuperPianoMan9167 (talk) 00:10, 19 November 2026 (UTC)
I actually think the current proposal would allow both. Samuelshraga (talk) 07:32, 19 November 2026 (UTC)
Related: WP:ORN#Contentious deletions and edit histories at two sidebar templates (authoritarianism during Trump, immigration policy of Trump) SuperPianoMan9167 (talk) 19:32, 18 November 2026 (UTC)
Given that both this and the almost simultaneous post to WP:NOR/N (which you strangely admit isn't a suitable place to bring it up) make multiple references to the editing disagreement you've had with Sirfurboy and just taken them to ANI over, this comes across as closer to WP:FORUMSHOP behaviour than you intended. If you want a genuine discussion on the issue have a single place where that's discussed instead of spreading it across multiple forums and hoping one vote ends up your way which is what this is starting to give the appearance of.
Also, once again as you've been warned about previously, stop editing your comments after people have replied. Rambling Rambler (talk) 19:34, 18 November 2026 (UTC)
One comment… if a nav template has grown to the point where it clutters up articles when presented as a sidebar… consider changing it to bottom box format. Blueboar (talk) 19:38, 18 November 2026 (UTC)
Many of these exist alongside the footers, or are larger than the fooder. Possibly the worst example, {{Conservatism US}} (every attempt to trim it is strenuously objected to), is somehow larger than {{Conservatism US footer}}. PARAKANYAA (talk) 19:45, 18 November 2026 (UTC)
Do we have articles that use both that sidebar navbox and that footer navbox? I would think it should be one or the other.
Shift the discussion away from arguments about the content of the navbox and focus it on placement. Sidebars should be small… Footers can be large. Blueboar (talk) 20:05, 18 November 2026 (UTC)
Sidebars should be small… I agree, but not everyone does. Most editors of these templates want them to be large. PARAKANYAA (talk) 20:22, 18 November 2026 (UTC)
I'm not sure whether editors "want them to be large" or if they "want them to be in a lot of articles, and then BIDI requires them to be large". WhatamIdoing (talk) 20:01, 19 November 2026 (UTC)
I haven't done a comprehensive check, but I think it's common for articles to have both. For example, the two templates on conservatism that Parakanyaa links have substantial overlap; in fact, one transcludes the other. And neither is BIDI-compliant, as they contain links to hundreds of articles, but are used only in dozens. WhatamIdoing (talk) 23:04, 18 November 2026 (UTC)
support' and think we should be more stringent. The community in general is moving away from the sidebars as they're being deleted on Mass..... Not as quick as there being made but still moving the right direction (in my view). We just need content editors to realize that the mass spamming of these is done with good intention....but our software will omit them from the mass majority of viewers.. We should make clear that this is a sort of see also spam run around is concern. Just because someone created or added a link doesn't mean the community has to accept 3 4 5 6 7 sidebars in one article. That should be reflected in our guideline. Waste of time for Content editors that have to deal with every inclusion or exclusion discussion when included in a sidebar supported by a guideline that promotes spam over prose text an educational imagery. Would support wording like we have with info box inclusion....that is that an addition to a sidebar doesn't automatically get included in an article simply because someone added a link in sidebar. Let's try to cut out the time sync that content editors have to deal with.Moxy🍁 19:47, 18 November 2026 (UTC)
That's another possibility for weakening BIDI in the name of reducing clutter:

An article should normally not transclude more than one or two sidebar navboxes, no matter how many sidebars link to them.

There's probably a better way to phrase this, but if we decide to create a series of small sidebars (e.g., a Taylor Swift discography sidebar, Taylor Swift filmography sidebar, Taylor Swift awards sidebar, Taylor Swift personal life sidebar, Taylor Swift tours sidebar, etc.), they does not mean that the Taylor Swift article itself needs to transclude all of them, or that they should be banned from linking to the Taylor Swift article. BIDI should not force including too many sidebars in any article, and BIDI should not ban inclusion of an obviously central link just because the sidebar isn't wanted at that central article. WhatamIdoing (talk) 23:45, 18 November 2026 (UTC)
Strongly Oppose the suggestion for what that's worth. All I see it resulting in is sidebars that are functionally unusable because they get turned into an inclusionist's wet dream of just a million "unidirectional" links to every article that is even tangentially related that no one then bothers to read on the few articles they are actually then placed, and more likely than not will see the sidebars themselves removed from almost every article because of all the space they take up like a crappy 2000s banner ad.
Between sidebars that are becoming articles unto themselves, see also sections that are extremely generous in argument as to how things are related, and external links sections that are borderline advertisements for websites the desktop Wikipedia experience increasingly feels like it needs a bespoke adblock service just to be able to read the article. Rambling Rambler (talk) 20:03, 18 November 2026 (UTC)
I think the biggest complaint content editors have is that the side bars take up room where an educational image would be more suitable because sidebars aren't seen by the vast majority of readers. Perhaps the guideline could mention the fact that we have {{if mobile}} that is a work around so that content editors can display an educational image over random link span. Moxy🍁 20:22, 18 November 2026 (UTC)
To be honest, even as someone who abhors the "mobile first" internet we are now stuck in, I think the fact that is de facto the case should instead highlight maybe instead of continuing to see the endless waste of editor time on yet more arguments over how best to use sidebars it should instead be recognised that they are now effectively redundant (because almost no one uses them but the people wanting to put links in them) and in the medium-term should seek to remove them entirely as a now redundant item. Rambling Rambler (talk) 20:32, 18 November 2026 (UTC)
Option A: sidebars are now visible for mobile users and start collapsed (right side), while they start expanded (left side) by default for desktop users, and if there's clutter there's a new parameter you can use to force it to start collapsed regardless of platform; click button to expand/collapse
Option B: deprecate sidebars and merge all existing sidebars into navboxes, improve navbox visibility by moving them to "See also" sections as new norm, make navbox norm to be expanded by default (overridable), and merge sidebars into navboxes (new Wikipedia:MOS norms would need to be established) (exceptions could be made for navboxes like Authority Control Database, which could remain at the bottom of the page). [EDIT:] We could even have all sidebars start collapsed by default (on mobile and desktop) to save real estate, I wouldn't necessarily be opposed to this.
To help you better understand my perspective, the reason I think sidebars are superior to navboxes is due to how sidebars are seen by most desktop readers, whereas most readers have no idea navboxes exist or to even look out for them (I still remember stumbling upon my first navbox a decade ago and thinking "what the hell is this?").
I drafted two WP:BOLD proposals, let me know what you think ~ Alexandraaaacs1989 (talk) 21:03, 18 November 2026 (UTC)
The fact that sidebars are so visible is the problem with them. We cannot have 1/3rd of the screen on many of our most visible articles being taken up by a useless taxonomy box. PARAKANYAA (talk) 21:09, 18 November 2026 (UTC)
Making sidebars visible for mobile users won't work as they'll take up the entire width of the page (and push down the rest of the content) regardless of whether they're collapsed or not. SuperPianoMan9167 (talk) 21:09, 18 November 2026 (UTC)
Both are terrible, because they both constitute the same problem which is a proposal that boils down to "let's stuff even more links into endless list of links no one will want to read because they're endless lists of links". You seem inherently inclined on every aspect of Wikipedia to believe the problem is that there isn't enough text already in existence and that somehow eventually if you add enough text it will solve every issue.
Wikipedia is meant to be an encyclopaedia, something that is easily accessible and gives you the summary of something. All this does is make achieving that harder and increases friction. Rambling Rambler (talk) 21:58, 18 November 2026 (UTC)
Can you please be WP:CIVIL/WP:NPA? ~ Alexandraaaacs1989 (talk) 22:09, 18 November 2026 (UTC)
You asked me what I thought and I gave you my view. None of my comment was a personal attack, but if you'd like a personal comment maybe notice how once again you've started responding to practically everyone in a discussion and dominating it (mere days after an ANI where the closer specifically gave you warning about this).
I said I didn't support it and why. Shoving two infoboxes effectively reiterating something that I've already said I strongly opposed as a reply to me is tantamount to WP:BLUDGEON. Rambling Rambler (talk) 22:40, 18 November 2026 (UTC)
Take it to my talk page, or take it to an ANI. Just stop yelling at me here ~ Alexandraaaacs1989 (talk) 23:33, 18 November 2026 (UTC)
Gordon Bennett, a handful of words bolded for emphasis is not yelling... Rambling Rambler (talk) 23:50, 18 November 2026 (UTC)
When these were first omit fom mobile view I remember there being a discussion about having and "Editor skin" (a simple add-on) that would allow these to be displayed to registers users who opt in by way of a tool or gadget ...a preference thing..... regardless if they're on mobile or desktop view. The discussion never really evolved because of data overload concerns on mobile view I guess..... But I'm wondering if someone out there can make us a skin so that there's an editor mode so that we always see them? Moxy🍁 22:07, 18 November 2026 (UTC)
Strongly oppose a change that would remove the only meaningful control we have for the over-large synthetic visually intrusive creations. This is the wrong question. SIDEBARS need to be smaller, with fewer links. Per Rambling Rambler, the effect of this will be to turn them into complete taxonomies, which means they will be monstrously intrusive on those pages they do get included on. Anything will then be added and any removals will be resisted. These consume way too much time as it is, and this change will exacerbate the problem further. Sirfurboy🏄 (talk) 20:53, 18 November 2026 (UTC)
what's horrible is the community has to waste so much time on these despite the fact that they're not seen by 80% of our viewership. The foundation has recognized that these are inaccessible and a data overload that are not beneficial overall. Moxy🍁 21:10, 18 November 2026 (UTC)
What's the context for the 80% figure? — Rhododendrites talk \\ 22:24, 18 November 2026 (UTC)
Just rounding up some stats "76.32% of all traffic to Wikipedia originates from mobile devices, while only 23.68% comes from desktops".Moxy🍁 22:42, 18 November 2026 (UTC)
Oh it's just mobile traffic? We have that for Wikipedia -- it's closer to 60% as of last year and ... apparently trending in the other direction somehow? But also, somehow I didn't know sidebars didn't appear on mobile! The More You Know! — Rhododendrites talk \\ 22:55, 18 November 2026 (UTC)
Our stats are very interesting.... they separate the app views instead of lumping them in together with mobile views. I believe many are unaware of this fact...that is they're not seen in mobile view and the app and that's why there's so many disputes over people just implementing {{if mobile}} Moxy🍁 23:01, 18 November 2026 (UTC)
My cynicism would say that the reason it's going up for desktop is AI related bots and scraping. Rambling Rambler (talk) 23:04, 18 November 2026 (UTC)
That was my thought, too, and then I forgot it by the time I actually posted the above. :) — Rhododendrites talk \\ 23:07, 18 November 2026 (UTC)
I agree with @Sirfurboy that SIDEBARS need to be smaller, with fewer links. WhatamIdoing (talk) 23:34, 18 November 2026 (UTC)
The problem I have with sidebars is they should require a much broader consensus for creation than they currently do, and often involve a nontrivial amount of OR and/or POV. BIDI should apply, and all of the articles that would be included should be notified of its proposed creation (and inclusion of that article). As it stands, they have a bizarre kind of implied framing authority based on [usually just one or two] editors' sense of "relatedness". I'm guessing this proposal has a background in OP's Template:Authoritarianism during the Trump presidencies sidebar? There's certainly consensus to characterize a broad swath of Trump's actions and style as authoritarian, but in many of the linked articles, the template is the only place the word "authoritarian" appears in wikivoice, and the grouping seems more like "Trump scandals the template creator considers typical of authoritarianism" rather than something rigorously sourced. So, on the whole strong oppose as moving in the wrong direction. — Rhododendrites talk \\ 22:34, 18 November 2026 (UTC)
@Rhododendrites, imagine a fairly simple sidebar. The OP mentioned {{LGBTQ sidebar}}, so consider the much shorter, simpler version I've mocked up here. It has about 20 links in it (vs. ~225 in the original). Can you imagine this being appropriate for more than these 20 articles, e.g., as a quick way to get back to the main/most popular articles on this subject? My desire to not have literally hundreds of articles in sidebars is why I think that BIDI needs to be relaxed for sidebars. WhatamIdoing (talk) 23:23, 18 November 2026 (UTC)
What are some examples of articles that should have this sidebar, but should not be included in the sidebar? — Rhododendrites talk \\ 23:57, 18 November 2026 (UTC)
Demiromanticism, Androphilia and gynephilia, and Aegosexuality seem like some good examples ~ Alexandraaaacs1989 (talk) 00:15, 19 November 2026 (UTC)
We could also have a [[Special:WhatLinksHere/Template:LGBTQ sidebar|What links here?]] What links here? hyperlink at the bottom of non-BIDI sidebars ~ Alexandraaaacs1989 (talk) 00:28, 19 November 2026 (UTC)
Why would we need to include this in the sidebar itself if anyone who wants to see what links to it can already do so by going to the sidebar page? SuperPianoMan9167 (talk) 00:31, 19 November 2026 (UTC)
For less technical readers :) ~ Alexandraaaacs1989 (talk) 00:38, 19 November 2026 (UTC)
Support. Practically, strict bidirectionality is the cause of navbox bloat - see WhatamIdoing's comments above. Anything loosening it will help us reduce the size of overburdened navboxes with too much going on.Katzrockso (talk) 23:29, 18 November 2026 (UTC)
Oppose, strict bidirectionality is not what is causing sidebar bloat. At a random look at the sea of blue that is {{LGBTQ sidebar}}, I saw the arbitrary collection of "Violence against LGBTQ in X" entries, with a short number of entries running from Belize to the US. The sidebar is in none of these articles. Some are not even articles; the sidebar is being used to hold Redirects to sections. CMD (talk) 23:58, 18 November 2026 (UTC)
I think this is a wonderful point but I'm also thinking many think things have to be done in baby steps. Moxy🍁 04:14, 19 November 2026 (UTC)

Proposal: Sidebar navboxes should be short

What do people think about this?

In principle, sidebar navboxes should contain a limited number of links. No matter how many pages they are placed on, and even if sections are collapsed by default, a sidebar should normally not contain links to more than about ____ pages. Longer collections of links should either be cut, re-formatted as footer navboxes, or split (e.g., an oversized "Example sidebar" with sections on Foo, Bar, Baz, and Bat could become a shorter "Example sidebar" with general links for use on relevant pages, plus separate "Example foo sidebar" "Example bar sidebar", "Example baz sidebar", and "Example bat sidebar" templates, each for use on a smaller number of pages relevant to that subtopic).

What number would you put in the blank? (I really think a number is necessary, no matter how hedged and approximate, or we'll have someone saying that hundreds of links is "short" because there are thousands that he could have included.) WhatamIdoing (talk) 21:32, 19 November 2026 (UTC)

No matter how many pages they are placed on implies links will not be bi-directional. If the sidebar is transcluded on a page, it should be linked from the sidebar. Anything else removes WP:BIDI by the back door. As for the number of links... hard to say. I'd like to see SIDEBARS in the tens. The existence of long established sidebars with a hundred or more links, however, will probably make anything like that hard to achieve. If we set the number at 100 or more, there will be little incentive for people to aim for something more compact. Sirfurboy🏄 (talk) 21:49, 19 November 2026 (UTC)
We should set the number at say 15. Yes, there are many "long established sidebars with a hundred or more links", but that is far too long, and they should be gradually made much smaller. Johnbod (talk) 22:25, 19 November 2026 (UTC)