Wikipedia:Signs of AI use in comments

This is a list of writing and formatting conventions typically found in comments written by Wikipedia users who are (or have been) suspected of using large language models or similar technology to write content, with real examples taken from talk pages and other discussion pages. Comments suspected of having been pasted from an LLM may be collapsed via {{collapse AI}} per WP:AITALK. Using AI to write comments is considered disruptive, and editors who do so may be blocked if they persist after being warned.

Although several of the tells mentioned at Wikipedia:Signs of AI writing (including boldface, em dashes, curly apostrophes and quotation marks, negative parallelisms, vertical lists, Markdown, and the rule of three) often appear in such comments, this project page only includes tells that typically do not appear in content added to articles or drafts. However, language that is overly procedural or spells out Wikipedia's core content policies may also appear in edit summaries.

Moreover, this list is descriptive, not prescriptive; it consists of observations, not rules. Advice about formatting or conduct can be found in the talk page guidelines, but does not belong on this page.

Content

Overassurance of quality or goodwill

Most people on Wikipedia want others to believe that they're here for the right reasons and are willing to follow the rules, and may assert that they do make their contributions with such rules in mind. AI chatbots, however, have a strong tendency to communicate this in a specific way: invoking policies, guidelines, and standards as a broad, formal, and abstract whole, as in legalese, and often using "AI vocabulary" to do so.

Examples

Regarding the ongoing AfD discussion, I respect the community process and will not contest it by continuing to recreate or resubmit articles that are under discussion. My removal of the AfD template was a mistake on my part, and I now understand that it should not have been removed.

[...]

My intention has always been to contribute constructively and improve Wikipedia's coverage of Malaysian elections and constituencies. I appreciate the feedback from everyone involved and will work within the community's expectations going forward.

— From this July 2026 ANI comment

Links to searches (many of these are common in unblock requests)

Policy-oriented:

Denial of promotional intent

When a user is suspected of engaging in conflict of interest editing, or their contributions are identified as promotional, they may deny that their edits are promotional and assert that they're just trying to share information about a particular subject. Such statements typically come in the form of negative parallelisms of the "not X but Y" variety. If writing about a company, they would express a desire to provide information about that company's offerings.

Examples

This page is not unambiguously promotional under Wikipedia’s criteria for speedy deletion (A7/G11).

The article is written in a neutral, encyclopedic tone and does not contain advertising language, calls to action, or subjective praise. All statements about the company are factual and are supported by citations to independent, reliable secondary sources from recognized news outlets.

The purpose of the article is to provide verifiable, encyclopedic information about the company (such as background, activities, and publicly documented presence), consistent with how similar companies are described on Wikipedia. It is not intended to promote the company, but to document it in an informational and neutral manner.

— From this January 2026 comment at User talk:567writes

I would like to clarify that my intention is not to use Wikipedia for promotional or advertising purposes. The article was created to provide factual and encyclopedic information about me, my professional background, work, achievements, and contributions.

[...]

My objective is not to promote any business, brand, service, or commercial activity through Wikipedia. I only wish to have an appropriately written, neutral, and well-sourced article about my professional journey and publicly documented work, provided that it meets Wikipedia's notability requirements.

— From these identical August 2026 AfC help desk comments by different users

Links to searches

Dubious assurance of human curation

When called out for using an AI to write or rewrite their content, some users will admit to using a large language model or similar technology to write or rewrite their content, but then follow that up with an assurance that they have checked their content to ensure it aligns with Wikipedia's editorial guidelines, or that the words in their comments truly reflect their thoughts.

Examples

I have reviewed WP:AITALK, and I am fully aware of the guidance. My comments here represent my own editorial judgment and responsibility. Suggesting otherwise is not appropriate.

— From this February 2026 comment at User talk:Grayfell

Fair one, I'm not trying to dismiss this. I am an intermediate English speaker. I used Grammarly’s editing tools (including grammar checking and clarity suggestions) to ensure my formal AfD nomination followed encyclopedic prose standards. Per WP:AI, using AI assistance for language and formatting is permitted provided factual claims are verified by the editor. I manually verified every source and citation in my nomination as evidenced by my correction of the URL error when I discovered it during source-checking. The substance of the nomination stands regardless of the drafting assistance used.

— From this June 2026 comment at Wikipedia:Articles for deletion/Rockstar (2026 film)

On the AI question? yes, I’ve used AI tools, but in a specific and within the policies way. I’m Spanish spoken, English isn’t my first language, and I’ve used AI to help with grammar, phrasing, and to check the structure of longer pieces. I haven’t used it to generate the actual content. Every fact, every citation, I’ve gone back to the source material myself and verified it before publishing.

Also, just to clarify how I work and why the text flagged, I do all my actual research and writing in Spanish first. Since English isn't my native language, I just run my finished text through a basic translator to pull it over into English. The ideas are 100% mine, but translation software naturally uses very predictable phrasing patterns, which is exactly why those AI detectors flagged it.

So to clarify the talk page posts, as well my process, is that I think and write out my thoughts in Spanish first. Since English isn't my native language, I can make some grammar errors, so I run my finished text through a basic translator to post it here. The actual thoughts and questions, are mine, but because standard translation software uses predictable phrasing patterns, it can end up looking like AI generated text to detector tools.

I want to explain how I work to answer some of the questions about automated text. I am not using a Large Language Model (LLM) to write my posts or generate content. My native language is Spanish, which is the language I use to do my research, map out my thoughts, and write my initial drafts. Because English isn't my native language, I just use a basic translation tool to turn my own finished writing into English so I can share it here. The predictable phrasing patterns that naturally happen with standard translation tools are exactly what triggered the automated AI detectors. [...] The research, ideas, and facts in my edits are entirely the result of manual human effort.

I am nearly fluent in English, but as a technical writer, I choose to use standard drafting and translation tools specifically to ensure that the structure, clarity, and paragraph flow meet the right standard for the reader.

Using tools to help organize ideas, check grammar, or improve structural formatting is entirely permitted under WP:LLM. It is also VERY discriminatory to weaponize WP:CIR against an editor simply because English is their second language, especially when the underlying research, facts, and citations are completely verified by human effort.

I am not disregarding the guidelines, nor discussing in bad faith, I am applying them exactly as they are written. The facts, research, and citations in my article are 100% verified by human effort.

— From all of these July 2026 comments at the AI noticeboard

Assertions of notability

Whereas AI-generated articles emphasizes a subject's impact, their creators' comments may exaggerate the extent to which a topic meets Wikipedia's notability guidelines.

Examples

The Pakistan Pharmacist Federation (PFF) is a well-established national organization in Pakistan, representing a significant and influential body of professionals in the pharmaceutical sector. The Federation plays an essential role in advocating for the pharmacy profession, advancing pharmaceutical education, and improving public health through various initiatives. Given its established presence, contributions to policy-making, and direct impact on the healthcare system, PFF qualifies as a notable entity under Wikipedia’s notability criteria for organizations.

[...]

Given the Pakistan Pharmacist Federation’s notability as a national organization with a significant role in public health and professional advocacy, and my commitment to revising the article to ensure it complies with Wikipedia’s editorial standards, I respectfully request that the speedy deletion nomination be reconsidered. I am confident that once revised, the article will meet Wikipedia’s requirements and provide valuable, neutral information about the Federation’s role in Pakistan’s pharmaceutical sector.

— From this May 2025 comment at User talk:Tahanazir

Over the past several months, I have worked to address the concerns raised during the review process. The draft includes significant coverage from reliable, independent sources that are not connected to the subject. Based on Wikipedia's notability guidelines, I believe the available sourcing demonstrates that the subject meets the criteria for inclusion.

Lovern is a notable figure within the comic book industry, particularly for his pioneering role in bringing computer colouring to comics through his company. Given the scope of his contributions and the available independent coverage, I am finding it difficult to understand why the article continues to be declined.

— From this June 2026 AfC help desk comment

Uncertainty about content issues (and requests for input)

When a user sees that an article they added content to has an {{AI-generated}} tag on it, or a draft they have worked on is declined or tagged for speedy deletion for being AI-generated, or they are accused of using AI to write their comments, they often act as though they have no clue as to what is wrong with their content. They may request additional input or assistance to help them understand exactly where they are going wrong and how they can improve. They want to know precisely which passages are causing concerns so they can make edits to address those concerns.

Canned expressions of willingness to improve one's work or take constructive criticism for edits are very common in AI comments by users whose drafts were declined by reviewers, or in rare cases, themselves. The difference between this sign and similar reassurances by conscientious humans will generally become obvious after you actually provide that criticism.

Examples

I understand that the AfC queue is heavily inundated with low-quality machine-generated content, but could you please point out which specific sections or phrasings raise concern? Alternatively, if the core issue lies in establishing the standalone notability (WP:GNG) of the term "digital patent" rather than the prose style itself, please let me know so I can locate more extensive secondary coverage or merge this concept into an existing broader article (such as Tokenization (blockchain) or Patent).

Thank you for your time and guidance.

Which specific model or detection method are you relying on to assert that this submission was generated by an LLM?

'AI-generated' is not, in itself, an official or actionable rejection criterion under Wikipedia guidelines. If this article fails to meet core standards, please specify precisely which policies are being breached: is the issue Notability (WP:N), Verifiability and Reliable Sources (WP:V / WP:RS), Neutral Point of View (WP:NPOV), or Manual of Style (WP:MOS)?

[...]

I am a human editor, not an LLM. If there are factual or stylistic deficiencies in the draft, I am open to constructive feedback. If you have valid, policy-based justifications for your decision, I would appreciate a clear and substantive response to these points.

— From these September 2026 comments at User talk:CoolIger5


Links to searches

If [there are / you have] (any) specific:

Canned requests for third-party review

When confronted by criticism of their work, it is common for LLM editors to make a stock request for an "experienced," "uninvolved," or "neutral" editor or administrator to provide a second opinion of their writing.

Examples

Hi, it's been 2 years since the warning banner has been placed and the article has been improved significantly. Would an independent editor please review the article to see if the banner is still appropriate? If not, would someone please remove the banner? Thank you!

— From this May 2026 comment at Talk:European Science Foundation

I respectfully request that an administrator:

  • Review the article and its sources.
  • Provide guidance on what specific changes would be needed to address the concern.
  • Restore the article and template if they are found to be valid contributions.
— From this June 2026 ANI comment

Before resubmitting through AfC, could an uninvolved editor please check out the current draft and advise whether the LLM maintenance tag can now be removed or if anything further need be addressed?

— From this September 2026 AfC help desk comment

When a draft is declined, LLM users tend to request source assessments in a similar manner. This usually carries the (unintentional?) implication that the reviewer didn't do their job properly.

Links to searches

Confusion over the reason for a declined draft

AI chatbots cannot read the decline notice left on the draft, and can only respond to whatever information regarding the declined draft that the user gives them. As a result, AI-generated questions regarding declined drafts will often express uncertainty or request clarification over the reason a draft was declined, even if the decline notice on the draft is very clear.

Examples

Could you please provide more specific feedback on the reasons for the decline? In particular, I would like to understand: Whether the issue relates to notability, sourcing, or tone If any specific sections need revision or removal What changes would make the article suitable for acceptance I’m committed to aligning the article with Wikipedia’s standards and would greatly appreciate your guidance.

— From this April 2026 AfC help desk comment regarding Draft:Ismatzein, which was said to have been promotional and failed WP:NBOOK

Is the primary issue related to notability, sourcing, tone, conflict of interest, or article structure?

— From this May 2026 AfC help desk comment regarding now-unavailable revisions to User:Revathyradhakrishnan2412/sandbox, which contained AI-generated text that went into detail about WP:CORPROUTINE activities

I would like some feedback if the issue is tone, sourcing, potential close paraphrasing, or article structure.

— From this June 2026 AfC help desk comment regarding Draft:Ontario employment Equity Act, which was declined for appearing to contain LLM-generated text

Is the concern related to factual accuracy, source use, writing style, article structure, or another issue?

— From this June 2026 comment at User talk:Helpful Raccoon regarding Draft:Digital Capital, which was declined for appearing to contain LLM-generated text

They may even compose self-contradictory comments that simultaneously acknowledge the specific reason for the draft's decline, but go on to say that they are trying to understand whether that was really the reason, or whether it was declined for some completely unrelated reason.

Links to searches

Dismissal of AI usage accusations as unfounded

When a user is confronted for allegedly pasting content from a large language model, they will likely dismiss these concerns as unsubstantiated speculation based on tone or writing style rather than hard evidence of LLM use. Most if not all users who do so also tell those who confront them to just point out what needs to be improved and focus on that instead of calling them out.

Examples

“Reeks of LLM” is pure speculation. You may believe that statement strengthens your argument, but in practice it weakens it.

Quite frankly, unsupported speculation and subjective interpretation can be more problematic than AI itself, because at least AI concerns can be evaluated through evidence, sourcing, and verification standards.

The issue is not whether LLM-generated article content is prohibited. The issue is whether conclusions are being drawn based on evidence or based on personal impressions and emotional interpretation of writing style.

Until consensus is reached through proper discussion and policy-based reasoning, I will continue removing warnings and templates that I believe are unsupported, speculative, or improperly applied.

— From these May 2026 comments at the AI noticeboard

Absent specific issues, a general impression about how the text was produced is not actionable.

You've just confirmed that your entire basis for this accusation is a personal impression formed by reading text. That's precisely what WP:NOLLM states is insufficient grounds for sanctions: "evidence beyond basic stylistic or linguistic indications." You have no method. You have a hunch. That's not actionable, and it never was.

Just look at how much time you spent on this exchange. It's time that could've been used to improve an article or do something else productive. The conclusion is, you just read a few words and formed an impression. That's all this ever was.

And no, I'm not going to answer a question built on an unsubstantiated premise.

As for conduct, I have engaged with every substantive point raised in this exchange. What I have not done is answer a question built on an unsubstantiated premise, which I am under no obligation to do.

More importantly, you have spent this entire exchange failing to identify a single factual error, a single unsourced claim, or a single content failure in the edit in question. Everything else has been an impression, an accusation, and now a hollow warning. The record of this conversation speaks clearly for itself.

Identifying a stylistic pattern is not evidence of AI generation, it's the same subjective impression restated with a grammatical label.

— From all of these June 2026 comments at User talk:Galilean-moons

Requests to focus on content quality instead

In many cases, editors have responded to accusations of AI use by suggesting that the question of how their content was made is not as important as the question of whether their content adheres to Wikipedia's other policies and guidelines. These statements often take the form of negative parallelisms such as "Let's keep this focused on X rather than Y", "Instead of X, this discussion has veered off into Y" or "the correct course of action is to X, not Y".

When told that the general pattern of their editing has been disruptive, they often dismiss the accusation as being merely speculative, unfounded, or having no basis in policy, and accuse their critics of acting in a manner that they believe is uncivil, hostile, dismissive, unprofessional, counterproductive, or an affront to Wikipedia's collaborative nature. Persistence from others is interpreted as aggression; if challenged sufficiently they may state their intent to withdraw from the discussion as it has become "emotionally charged". Since the editor did not write this comment themselves, they will often continue the discussion or return to it shortly afterwards.

AI will seldom confirm that it's being used without the prior consent of the user, so if the response is vague, sidesteps the question or considers it to be a personal attack, this may be a sign of AI use.

Examples

Highlighting key:

  • "Instead of making accusations, let's focus on fixing this article."
  • "Let's focus on fixing this article instead of making accusations."
  • "I consider your accusations to be frustrating and inconvenient."
  • "The correct course of action is to help me fix this article, not to make accusations."
  • "Instead of helping me fix this article, you kept making accusations."
  • "You have continued to make accusations instead of helping me fix this article."
  • "The correct course of action is to help me fix this article. Making accusations is a bad idea."
  • "This discussion has shifted from productive discourse to accusatory remarks."

If there are particular words or phrases in the article that make it appear AI-generated to you, wouldn't it be more helpful to simply point them out and help correct or improve them?

Wouldn't that be more productive than repeatedly nominating the article for deletion based on a suspicion about how it was written?

[...]

If you believe the article violates a specific Wikipedia policy, please identify the specific policy and the specific content that you believe violates it, rather than speculating about whether AI was used.

I would much rather work with you to improve the article than have it repeatedly nominated for deletion based on an assumption that I have already explained is incorrect.

— From this October 2026 comment at Wikipedia:Articles for deletion/Bajestan Playa

Misrepresentation of Wikipedia's policies and guidelines

When AI chatbots discuss Wikipedia's policies or guidelines, they may provide incorrect or outdated information about them. In several cases, they have misattributed a rule to a shortcut whose target doesn't contain it at all.

Examples

WP:NOTENGLISH clearly states:

"Wikipedia articles should not contain material based solely on non-English-language sources without reliable English-language sources."

In this case, the material you’re suggesting must be accompanied by a reliable English translation or reliable secondary sources for verification. Without that, it cannot meet Wikipedia’s verifiability standards.

— From this April 2025 ANI comment

Per WP:LAW: "Wikipedia is subject to the laws of the countries in which it operates."

— From this August 2025 revision to User:Flanker235
  • WP:BLOCKPOLICY — blocks are preventative, not punitive.
  • WP:BEFORE — admins should attempt communication before blocking.
  • WP:AGF — assume good faith.
  • WP:BITE — do not bite newcomers.
  • WP:INVOLVED — avoid actions where involved.
  • WP:UNBLOCK — blocks may be overturned when not aligned with policy.
— From this December 2025 unblock request

In some cases, while citing a guideline, an AI chatbot may state information that can be found at a related project page, but will fail to mention that other page and instead misattribute what it says to the guideline.

  Correct for WP:LLM
  Correct for WP:LLMS § Writing articles
  Incorrect anywhere

I want to make sure I understand the policy correctly though, WP:LLM does list permitted uses beyond spelling and punctuation, including using AI to “spot structural problems,” “generate ideas,” and ask “how to improve a paragraph.” My use was in that advisory space, with all prose written and verified by me. I’m not trying to argue in bad faith, I really want to understand where the line is so I can stay clear on the right side of it.

According to the policy documentation at Wikipedia, Large language models (WP:LLM), my workflow is explicitly protected. Under the section "Using LLMs as an advisory tool", the policy exactly permits editors to use digital tools to "suggest refinements to their own writing, and to incorporate some of them after human review, provided the LLM doesn't introduce content of its own." It states that asking "how to improve a paragraph" is a permitted function. As well, under the "Enforcement" section of the policy, it states: "The writing style of some editors may appear similar to the output of LLMs, meaning that the imposition of sanctions requires evidence beyond basic stylistic or linguistic indications." The guidelines straightforwardly command reviewing editors that "When evaluating possible LLM use, it is best to consider the full pattern of the editor's recent edits and whether the edits comply with core content policies."

— From these June 2026 AI noticeboard comments

Non-existent shortcuts

In some cases, users of AI chatbots have pasted text containing hallucinated shortcuts that do not redirect to any existing page at all.

Examples

WP:ANACHRONISM – The sovereign state of Pakistan was established in 1947. No political, cultural, or geographical entity called "Pakistan" existed before this date. Using this name for 5000 BC history is a textbook anachronism.

However, every argument I have made is based on Wikipedia policies (WP:SYNTH, WP:NPOV, WP:ANACHRONISM) and mainstream academic reality.

  • Contradictory Logic: The editor claims my edit is skewed by personal affiliation (WP:COIBELIEF), yet relies on personal YouTube observations to dismiss standard ecclesiastical terminology. [...]
  • Misuse of COI (WP:COIBELIEF): Religious background is explicitly not a conflict of interest, and adding documented internal self-designation does not violate WP:NPOV.

Minimization of Wikipedia's prohibition on AI use

When confronted about using AI to write content, some users may downplay, disregard, or deny Wikipedia's prohibition on the use of large language models to generate content, and claim that whether the content they generate adheres to Wikipedia's other content guidelines matters more than how the content was produced.

Examples

Please remember, Wikipedia allows tools like AI as long as the content remains factually correct and properly cited.

AI-assisted drafting is not prohibited. Per Wikipedia policy, the focus is on verifiability, not the tool used for drafting. Every fact in this article is supported by high-quality, third-party sources. Claiming "AI" as a pretext to delete documented financial records is a violation of WP:POINT.

— From this October 2025 comment and this March 2026 comment at Talk:Arthur Katalayi

I can confirm that Grammarly is the tool I use for proofing. I also think it’s important that others participating in this discussion recognize that Wikipedia does not maintain a formal prohibition on the use of AI-assisted tools. What ultimately matters is whether the content complies with Wikipedia’s core policies regarding verifiability, neutrality, sourcing, and accuracy.

Wikipedia’s guidance regarding AI focuses primarily on preventing hallucinated or fabricated information. It does not prohibit the use of AI-assisted tools altogether, particularly when the content itself remains verifiable, properly sourced, and subject to human review and editing.

— From these May 2026 comments at the AI noticeboard

Exhaustively procedural language

Comments by LLM users, especially those in which a user suggests or expresses willingness to make edits to a particular article, are often exhaustively procedural about the editing process and would go into detail about how that user has tried to (or would) follow the rules, or how a particular article or draft is (or can be edited to be) in compliance with the rules.

LLM users that mention or identify specific policies or guidelines often spell out Wikipedia's core content policies or—in rarer cases—misquote existing project pages or cite shortcuts that haven't been created.

Examples

I appreciate the guidance regarding notability, neutrality, and article creation. After receiving the initial feedback, I substantially revised Draft:Matthew Mettias to better align with Wikipedia's content policies. [...] It has been rewritten with the goal of presenting information in a neutral, encyclopedic tone supported by citations.

I have also made a concerted effort to ensure that factual statements are attributed to published sources where available. The draft cites independent coverage, [...] alongside primary sources where appropriate for routine factual verification. My intention is to satisfy Wikipedia's requirements for verifiability while distinguishing between independent commentary and primary-source information.

If the draft still falls short of Wikipedia's standards, I would sincerely appreciate specific guidance on which aspects require further revision rather than immediate deletion. I am committed to improving the article in accordance with community policies and welcome constructive feedback.

— From this July 2026 comment at User talk:Mikemerris

I understand that every edit must be based on reliable sources, carefully reviewed, and represent my own judgment as an editor. [...] I will make all edits myself, ensure that every contribution is supported by reliable sources with appropriate inline citations, and seek guidance through talk pages whenever I am uncertain. Rebuilding trust will take time, and I am prepared to do that through careful, policy-compliant editing. [...] If unblocked, I intend to demonstrate through my future conduct that I have learned from this experience and can contribute constructively while fully adhering to Wikipedia's policies and expectations. Thank you for taking the time to consider my request.

— From this July 2026 unblock request at the administrators' noticeboard

Links to searches

Overjustification of editing suggestions

When an LLM user proposes edits to (or asks for help improving) a particular article, they may attempt to justify their proposals by stating that their desired edits would make that article adhere more to Wikipedia's policies and guidelines; improve its tone, formatting or structure; or make it more useful or convenient for readers. Such statements may resemble forms of collaborative communication where AI chatbots tell users that performing a particular action on a given article would improve it in a specific way.

Examples

I agree that neutrality WP:NPOV is very important here. The article currently relies heavily on investigative reports that describe serious allegations, but it gives very little weight to the company's own position and the lack of any formal legal outcome. To make the article more balanced and neutral, I suggest the following small, sourced adjustments:

(This reflects that the activity is based on media investigations rather than any court conviction or chargesheet.)

(Again, this keeps the sentence accurate to the sources while noting it is an allegation, not a proven fact.)

The article currently says "Appin-linked hackers" and "former employees went on to found other hack-for-hire firms." To avoid implying that Appin itself continued the activities, a brief clarification could be added (e.g., "hackers linked to former Appin employees") so readers understand the distinction between the company and individuals who left.

Current wording gives the impression of active, ongoing court cases. In reality there are no current indictments or charges. Suggest changing to "has been the subject of legal actions and media investigations" (removing "ongoing") to be factually precise}.

These changes would make the article more neutral without removing any reliably sourced information. They simply ensure the reader sees both sides clearly, as the reviewer has already noted is important.

— From this April 2026 comment at Talk:Appin (company)

Links to searches

Language and grammar

"Spelling out" of Wikipedia's core content policies

LLM users sometimes mention the titles of a few of Wikipedia's core content policies together, often alongside their corresponding shortcuts, like this:

verifiability (WP:V), neutral point of view (WP:NPOV), and reliable sources (WP:RS)

Examples

Thank you for your question. Yes, I did use an AI tool to assist with parts of initial drafting process. However, I understand that Wikipedia requires all content to meet its core policies, including verifiability, neutrality, and no original research.

— From this April 2026 comment at Talk:Black Diamond Museum, Angul

With good faith I am opening this discussion with intent to review the current state of this article, specifically focusing on verifying its adherence to core content policies including Neutral Point of View, Verifiability, and No Original Research. Upon a thorough review and research based on the reliable sources, several facts of the article appear to rely on ambiguous framing, original synthesis, and assumptions rather than direct, verifiable facts from high quality sources.

— From this June 2026 comment at Talk:Pasi (caste)

Links to searches

Frequent use of certain words or phrases

Some words and phrases are commonly used in comments from users who are suspected of using AI for writing assistance. Many common words and phrases can be found in other "Words to watch" boxes, but this section is a catch-all for words and phrases that either aren't exclusively associated with any other sign, or aren't necessarily associated with any other sign identified on this page whatsoever.

Examples

I understand the importance of adhering to Wikipedia's policies, particularly regarding self-promotion and the inclusion of non-article pages.

— From this March 2024 comment at User talk:Bandreddi VijayPrasad

Links to searches

Concrete (as an adjective)

Users confronted for suspected LLM use often use the word concrete as an adjective, especially whenever they insist on being given "concrete examples" of text that needs improvement or "concrete evidence" that they used AI aside from "stylistic indications". When they don't get the "concrete" answers they demand, they dismiss the AI accusations against them as merely subjective speculation.

Note that humans have often used the word concrete in this sense when discussing the distinction between abstract and concrete things.

Without concrete examples, your concern cannot be evaluated in line with WP:V, WP:RS and WP:BURDEN. — From this April 2026 revision to Talk:House of Dust (architecture)
I think you're the one who's incapable of being an editor if you can't express concrete things, but rather general concepts that have little meaning. — From this June 2026 ANI comment
If there are concrete concerns, such as inaccuracies, sourcing problems, or wording that does not meet Wikipedia standards, please point them out so they can be addressed. — From this June 2026 comment at User talk:Galilean-moons

If there is a concrete sentence in the article that violates a core content policy like verifiability or neutral point of view, please point it out so I can address it ASAP.

I’d appreciate concrete feedback on any statements you believe do not meet Wikipedia’s standards, so I can improve the article on that basis.

— From these July 2026 AI noticeboard comments

"I am happy to..."

LLM users often say that they would be "happy" to make whatever changes are needed to improve their work or bring it in line with Wikipedia's standards if other editors tell them what needs to be improved first.

3. Community Consultation I am happy to discuss this process with the appropriate editors or administrators to ensure alignment with community guidelines.
— From this November 2024 ANI comment

If it helps, I’m happy to go through the article section by section and show where each claim comes from.

If you can point me to a specific passage or edit that concerns you, I’m happy to address it

I'm happy to manually polish the phrasing going forward so it reads as naturally as possible.

— From these July 2026 comments at the AI noticeboard

If the preferred process is to seek consensus first, improve a smaller sample of articles, or work in draft space before creating additional articles, I am happy to follow that process.

— From this July 2026 ANI comment

If there is actual evidence connecting these accounts then please provide it I am happy to look at it and respond to it But if the conclusion is based only on edit timing user pages and the fact that the articles concern Nigerian companies then I do not think that is enough to make an allegation of this kind

As for the Breet article itself I am happy to deal with the article on its own merits I can review the references remove anything that is not properly supported and rewrite anything that does not meet Wikipedia's standards

[...] I have reviewed the wording and references and I am happy to make any changes that are necessary to bring the article fully in line with Wikipedia's standards.

If there is a particular passage that someone believes was generated by an LLM or a particular reference that is considered non-existent or unreliable, please point it out specifically. I am happy to address it. But simply assuming that an article was generated by an LLM because of its writing style is not the same as establishing that it was.

— From this September 2026 COI noticeboard comment

"I respectfully..."

LLM users often describe themselves as "respectfully" requesting that certain actions be performed, or "respectfully" challenging attempts to clean up after their AI-assisted contributions.

I respectfully contest the proposed speedy deletion of Draft of Lives (2026 film).

[...]

I also respectfully disagree with the characterization of the draft as promotional. [...]

I respectfully request clarification of the evidence supporting that assertion, the specific speedy-deletion criterion being invoked, and the passages considered promotional.

— From this October 2026 Teahouse comment

Links to searches

Failed attempts to cite project pages

Some comments have simply cited "WP" where a shortcut to a project page would be. This often results in the construction "WP and WP".

Examples

Additionally, I have ensured adherence to WP and WP , with proper disclosures where applicable. — From this August 2024 Teahouse inquiry

Removing only those words would preserve the existing attribution while bringing the sentence into closer alignment with WP and WP.

— From this October 2026 comment at Talk:IQ Option

I respectfully request that administrators examine:

  1. Whether the comments and interactions comply with [[WP]] and [[WP]].
— From this October 2026 ANI comment

Links to searches

Style

Subject lines

Historically, some comments generated by AI chatbots have begun with text that appears intended to be pasted into the Subject field on an email form.

Examples

Subject: Request for Permission to Edit Wikipedia Article - "Dog"

Subject: Edit Request for Wikipedia Entry

Subject: Request for Review and Clarification Regarding Draft Article

— From this November 2024 comment at the AfC help desk

Subject: Concerns about Inaccurate Information

Subject: Behavioral issues and Wikihounding by User:Binksternet

Links to searches

Request for:

Division of text into titled sections

AI chatbots sometimes generate messages with text broken into sections with titles. In some cases, such text may either appear with Markdown or as plain text when pasted. They may resemble or include inline-header vertical lists.

Importance of Thorough Research

Wikipedia's content guidelines emphasize the importance of verifiability and reliable sourcing, [...]

Risk of Violating Wikipedia’s Policy on Biographies of Living Persons

Wikipedia's Biographies of Living Persons (BLP) policy is particularly relevant in this context. [...]

The Dangers of Blindly Cutting Content

Blindly cutting content, especially when it pertains to significant aspects of a person’s public life, contradicts the core principles of Wikipedia. [...]

The Role of Due Diligence in Editing

Editors have a responsibility to engage in due diligence when making significant changes to articles, particularly those involving living persons. [...]

— From this August 2024 comment at Talk:Eric Dick (lawyer)

Response to Article Concerns

Source Independence and Coverage

The article has been updated to address concerns about source reliability. [...]

Swiss Media Coverage

The limited coverage in mainstream Swiss media likely reflects the niche nature of iolite Capital and its international investment focus. [...]

AI-Generated Content Suspicion

The perception that the article may be AI-generated seems to have been driven by previously listed dead or weak sources. [...]

— From this May 2025 comment at Wikipedia:Articles for deletion/Robert Leitz

Misuse of section subheadings

Some LLM-generated comments contain level 2 or level 3 subheadings, or syntax intended to produce those subheadings, sometimes in Markdown, in which cases ## is used to denote each section.

Examples

In some cases, the syntax for subheadings appears in replies, where indentation prevents subheadings from forming.

Transclusion of article maintenance banners

When mentioning maintenance tags, AI chatbots often just write the names of such templates in curly brackets (e.g. {{Example}}), resulting in unintentional transclusions. These can be avoided by typing tl| between the opening pair of brackets and the template's name (e.g. {{tl|Example}}) so that each mention will instead appear like this: {{Example}}

Examples

Links to searches

Miscellaneous

Chat log disclosures

Some commenters have copied and pasted their chat logs onto Wikipedia, including their prompts, which may give away their unassisted writing style.

Examples

This page should not be speedily deleted because (Log in

Sign up You said: I have write an article Tajmul ALi Chattar which is an historical place of fenchuganj on wikipedia. but the wikipedia authority is trying to delete the page saying that this page is promotional or about someone who is close to me .write a long message to convience them to publish the page ChatGPT said: ChatGPT Certainly! Here's a detailed message you can use to address the Wikipedia authorities and make a case for the publication of the page about Tajmul Ali Chattar:

[...]

You said: they are saying that the deletion has been done under section U5 of the criteria for speedy deletion, because the page appears to consist of writings, information, discussions, or activities not closely related to Wikipedia's goals, pls convience them to publish the page ChatGPT said: ChatGPT Certainly! Here is a message you can use to address the deletion under section U5 and make a case for the page's relevance and inclusion:

[...]

You said: they are saying that This user page may meet Wikipedia's criteria for speedy deletion as a page in userspace consisting of writings, information, discussions, and/or activities not closely related to Wikipedia's goals, where the owner has made few or no edits outside of userspace. See CSD U5. pls convience them to publish the page ChatGPT said: ChatGPT Understood. It seems that the deletion was under section U5 of the criteria for speedy deletion, which is used for user pages that don't align with Wikipedia's goals. Here’s a detailed message to address this issue and make a case for the page:

— From this August 2024 comment at User talk:Tajmul Ali Chattar Fenchuganj

ChatGPT

Saved memory full

You said:

What does it mean

ChatGPT said:

Could you please clarify what you're referring to by "What does it mean?" — is it a phrase, message, image, or something specific you saw? Feel free to paste or describe it so I can give you an accurate explanation.

You said:

[User pastes text from the notice that their user page has been nominated for speedy deletion]

ChatGPT said:

This message means that your Wikipedia user page (User:NJ Sumbal) has been flagged for speedy deletion under criterion U5, which applies to pages that are not clearly related to Wikipedia’s mission.

[...]

You said:

My notable work mentioned in Wikipedia

ChatGPT said:

Thank you for the clarification! If your notable work is already mentioned in Wikipedia, that gives you a stronger case to contest the deletion of your user page.

[...]

You said:

Please write me

ChatGPT said:

Certainly! Here is a polite and clear message you can use when clicking the “Contest this speedy deletion” button on your Wikipedia user page:

— From this July 2025 revision to User talk:NJ Sumbal
You said: Write full details about how Moru come to South Sudan ChatGPT said: Here's a detailed account of how the Moru people came to South Sudan, based on oral traditions, linguistic studies, and historical migration patterns: — From this July 2025 revision to User talk:Ayawa ariko


Links to searches

Canned unblock requests

When a user is blocked, they might ask an AI chatbot to write an unblock request for them. Many AI-generated unblock requests tend to go something like this:

Dear Wikipedia Administrators,
I respectfully request that the block on my account be lifted. I acknowledge that my recent behavior has been in violation of Wikipedia's standards. My intention was to edit constructively, and I sincerely apologize for any disruption I may have caused. I have taken the time to carefully read and review all of Wikipedia's policies and guidelines, and I now understand the importance of adhering to them when editing. Moving forward, I am committed to ensuring that all of my future edits are constructive. If there are any specific concerns that need further clarification, feel free to ask me about them, and I will be happy to address them. I look forward to collaborating in a constructive manner. Thank you for your time and consideration.

In these requests, blocked users often try to assure administrators of their good intentions, express their willingness to communicate and collaborate in the future, and request feedback to help them improve their editing, all in a canned manner.

Links to searches

Requests for actions:

Responses by others:

Generic language

An AI chatbot prompted to compose an unblock request generally has little to no information about the blocked user's situation to work with besides what the user provides about it. Thus, any unblock request it generates may read as though it could have come from any blocked editor, not just the one taking credit for that particular unblock request. In many unblock requests, users appear to make generic promises to engage in constructive, policy-based editing.

Examples

If my account is restored, I will focus on making constructive, neutral, and policy-compliant contributions. I understand the importance of maintaining Wikipedia's neutral point of view, using reliable independent sources, and avoiding promotional language.

Thank you for considering my request. I appreciate your time and I am willing to follow all applicable Wikipedia policies going forward.

— From this July 2026 unblock request at User talk:Nikuonspace

Links to searches (see also WP:AIASSURE and WP:PROCEDURALTALK)

I have taken the time to ______ the rules:

Apologies for potential past damage

In many unblock requests, users seem to apologize for "any" disruptive behavior "may have" engaged in, as opposed to the particular forms of disruption they certainly have engaged in that led to them being blocked in the first place. These apologies imply that the user is uncertain as to whether their behavior was wrong, or whether they engaged in wrongful behavior at all, even if they had been told multiple times that their behavior was wrong before they were blocked.

Examples

I would like to express my sincere apologies for any violations I may have unintentionally committed. Upon reflection, I realize that some of my previous contributions may have inadvertently crossed the line into promotional or advertising behavior.

— From this May 2023 unblock request at User talk:Circle 360 Creative Solutions

[...] I apologize for any content I contributed that may have violated these rules.

— From this May 2024 unblock request at User talk:Marcinebu1984

Looking back, I realize that some of my edits may have appeared promotional, and I also understand that my username could be interpreted as representing an organization or brand.

— From this July 2026 unblock request at User talk:Nikuonspace

Links to searches

Canned ANI reports

Large language models have been used to write reports at Wikipedia:Administrators' noticeboard/Incidents, as evidenced by the presence of indicators that have appeared on LLM-generated text elsewhere on Wikipedia. These reports often present each issue as entries on a list or even separate sections divided by subheadings, whose titles are written in title case. Some of them also use boldface, or attempt to do so via Markdown.

Examples

1. **Account Age:** [...]

2. **Prior SPI Context:** [...]

3. **Exact Lead Quote Re-insertion:** [...]

4. **Active Cross-Page Disruption:** [...]

Sealioning

Sealioning is a disruptive practice where someone feigns ignorance and relentlessly demands that users provide answers or evidence to back up the statements they make. When targets express frustration over the sealioner's behavior, the sealioner criticizes the targets for responding negatively to them, even though they were just asking questions politely, or so they present themselves as polite.

Several AI users have resorted to sealioning when confronted about their AI use. They write long comments that sometimes present arguments as entries in bulleted lists or separated into multiple sections. LLM users often ask for examples of problematic text or stronger evidence than "stylistic indicators", and accuse their confronters of acting on unsubstantiated speculation and being uncivil when they don't get the answers they want.

Wikilawyering

Wikilawyering is a disruptive practice where someone selectively cites or interprets policies, guidelines, or perceived precedent as justification for their conduct, even if their interpretations go against the purpose of the policies or guidelines they mention. In many cases, users have emphasized their content's compliance (or overlooked their non-compliance) with certain policies and guidelines. This is especially the case for users of AI chatbots, which often generate text that affirms what the user may want others to believe, even if the points they present don't actually hold up against Wikipedia's policies and guidelines.

When an article is tagged as possibly containing AI-generated content, a user might try to defend it by asking accusers to point to specific passages causing concern or reassuring them that the content they've contributed is "neutral", "verified" by citations to reliable sources, and comprises none of the items described in Template:AI-generated, regardless of the amount of actual effort put in to ensure that such claims are true.

Common AI wikilawyering tropes include:

On a couple of occasions, AI users have claimed that Wikipedia's guidance on LLM use permits the use of LLMs to "spot structural problems", "generate ideas" or for help "improv[ing] a paragraph". Although these specific words do not appear at WP:LLM (which they erroneously cite when mentioning these exceptions), they do appear in this paragraph at Wikipedia:Large language models § Writing articles, which may be what they mean to refer to:

The ban does not cover indirect use of LLMs, which means that they can be used adjacent to editing. For example, they can help editors spot structural problems in long articles (such as: the same information being repeated in different sections, the infobox not agreeing with the prose, etc.), and to generate ideas for new or existing articles. If using an LLM as a writing advisor, i.e. asking for an outline, how to improve a paragraph, a critique of existing content, etc., editors should remain aware that the information it gives is unreliable.

Ineffective indicators

False accusations of AI use can drive away new editors and foster an atmosphere of suspicion. Before claiming AI was used, consider whether the Dunning–Kruger effect or confirmation bias may be clouding your judgement. Detecting LLM texts on the basis of style alone is not as easy as it seems, see WP:AIDETECTIVE. Here are some somewhat commonly-used indicators that are ineffective in LLM detection—and may even indicate the opposite.

See also

Notes

  1. ↑ The shortcut WP:LLMUSE was created on July 2026, three months after this comment was created.
  2. ↑ See these discussions from July and August 2026.