Wikipedia:Presumptive removal of AI-generated content

The use of large language models (LLMs) and artificial intelligence (AI) to rewrite or generate article content is largely prohibited on Wikipedia because such edits often violate several of Wikipedia's core content policies. When a user has been determined to have a history of AI use, a series of their contributions may be assumed to violate Wikipedia's policies and guidelines. In this case, edits to existing pages may be reverted without review, and articles where the editor is the only significant contributor may be nominated for a special proposed deletion process, LLMPROD, subject to the requirements below. These two processes are collectively known as presumptive removal.

Any editor can reverse a presumptive removal. When doing so, they must review the content and its sources to ensure that it complies with policies and guidelines, which may require reworking the content.

Requirements

The following requirements must apply for presumptive removal of AI-generated content:

  1. The editor who added the content either:
    1. is blocked for AI use,
    2. is inactive and there is strong evidence of habitual LLM use,
    3. discloses that they have used AI and consents to clean-up of their past edits,
    4. or is subject to consensus at a noticeboard such as WP:AINB or WP:ANI that presumptive removal is appropriate.
  2. If a block remains active, it is not being appealed or discussed, and it is old enough such that the blocked editor has had sufficient time to appeal.
  3. The contributions to be removed coincide in timing and pattern with edits determined very likely to be AI-generated.

Process

For clarity, editors engaging in presumptive removal (whether through reverting edits or proposing articles for deletion) should include in their edit summary:

An example of such an edit summary may be:

Presumptive removal of LLM-generated content, see AI cleanup investigation. Feel free to reinstate by following the procedures for disputing presumptive removal of LLM-generated content.

This summary can be produced using the following code:

[[WP:LLMPRV|Presumptive removal of LLM-generated content]], see [[WP:AI noticeboard/User:Example|AI cleanup investigation]]. Feel free to reinstate by following [[WP:LLMPRVOBJ|the procedures for disputing presumptive removal of LLM-generated content]].

A justification specific to the article is optional but recommended.

When proposing the deletion of an article ("PRODing"), make sure the creator is its only significant contributor. Then, add {{subst:prod llm}} to the top of the article to nominate it. If the tag has not been removed within 5 days, an uninvolved administrator will check that the PROD complies with § Requirements and, if so, delete the article.

Reversing removal of content

An editor objecting to the removal of content may assume responsibility for the content, as well as the onus to achieve consensus for its inclusion. Assuming responsibility requires that all content and its sources be closely reviewed and reworked as needed, to ensure that they comply with policies and guidelines. Besides potentially removing the content, a review may involve checking that the cited sources support the content and are reliable, determining whether the content is due for inclusion, and reworking the content to revise the tone and resolve possible copyright violations. Editors may find 'stubbifying' to be the easiest solution, so long as what remains is reviewed. After these steps have been taken the reviewed and revised content may be restored. If objecting to a proposed deletion, also remove the {{prod llm/dated}} template from the page.

If the PROD timeframe is too short, editors are free to 'userfy' PRODed articles and remove the tag, provided that they do the above before moving them back to mainspace. If the article has already been deleted, any editor may request its undeletion, provided they also do the above shortly thereafter. Repeatedly restoring content that violates policies and guidelines will likely be met with sanctions.

See also

Notes

  1. ↑ RfC on presumptive removal of AI-generated content
  2. ↑ Including partial blocks.
  3. ↑ Defined as having not made more than 5 mainspace edits in the last 30 days.
  4. ↑ This may vary depending on user activity, but is usually around one week.
  5. ↑ It is necessary to determine when the editor first started using AI to generate content. Edits before 30 November 2022 (the release date of ChatGPT) are very unlikely to be AI-generated.
  6. ↑ An article can be LLMPRODed multiple times, contrary to the standard WP:PROD process wherein articles can only be PRODed once.