Wikipedia:Reverting AI edits
 | This is an essay on the Writing articles with large language models guideline. It contains the advice or opinions of one or more Wikipedia contributors. This page is not an encyclopedia article or a Wikipedia policy, as it has not been reviewed by the community. |
Since March 2026, the English Wikipedia community has virtually prohibited creating mainspace content with large language models, also known as AI chatbots.
Editors should not be expected to preserve content generated in large part by an AI chatbot. While an editor using such a chatbot is likely to be editing in good faith, it is well documented that AIs all too-often create convincing sounding hoaxes, introduce fictitious information or references, and misrepresent sources. It can take a long time to fact check them, and an even longer time to correct the edits. As such, editors may choose to summarily revert any substantial edits made by another editor with a confirmed history of AI abuse or misuse. This is especially true if the editor in question has been blocked or sanctioned for AI or LLM misuse or if the edits contain obviously unreviewed AI generated text such as communication from the chatbot or fictitious/implausible references.
To avoid collateral damage, it is advisable for editors to take reasonable steps to ensure that the series of edits in question is very likely to be AI-generated, and not compliant with Wikipedia's other policies and guidelines, before presumptively reverting. Some types of edits may be less problematic than others: for example, there may be very little need to revert an AI-generated infobox if its content is verified in the article. Some (but not all) AI copyedits are either genuine if tiny improvements, or so inconsequential that they're hardly worth the time.
On the other hand, a completely AI-generated rewrite of an article, with tone issues, hallucinated references, or original research, should be reverted to the last, non-AI-generated revision. Editors ought to attempt to preserve any reliable sources, obvious corrections, or unrelated constructive edits by other people, if reasonable, but they are expected to use their best judgement.
Any editor can take responsibility (WP:ONUS) for content that is presumptively removed, though they should check and edit it to ensure it complies with Wikipedia's policies and guidelines.
Special cases
[edit]BLP challenges
[edit]If an editor, even one using AI, makes a good-faith argument that part of an article contains material about living persons that is poorly sourced, or even not sourced at all, please do not revert and re-introduce that material. Compliance with the Biographies of living persons policy takes precedence over LLM-related concerns.
This does not mean that you are not allowed to revert somebody claiming that article text may be defamatory. However, you should investigate their claims. When in doubt, you can remove the AI generated material, but please avoid reverting back to the questioned material until you can confirm it is adequately sourced, neutral, and a fair reflection on the sources. You may ask for assistance at the Biographies of living persons Noticeboard.
Likewise, if you are an article subject, and somebody has reverted your edits, citing concerns over (perceived) AI use, please feel free to raise your issues with the article at the Biographies of living persons' Noticeboard.