Wikipedia adopts presumptive removal of AI-generated edits

Wikipedia has quietly formalized something that feels like a real turning point for how the web's most-used reference deals with the flood of AI writing: contributions from editors with a demonstrated history of AI use can now be wiped without individual review. The guideline, Presumptive removal of AI-generated content, was last updated on 22 July 2026 and describes the use of large language models "to rewrite or generate article content" as "largely prohibited."

The mechanics are worth understanding, because they represent a shift from evaluating content on its merits to evaluating it by the editor's history. Under the guideline, contributions to be removed must "coincide in timing and pattern with edits determined to very likely be AI-generated." Removal proceeds only if the editor is already blocked for AI use, is broadly inactive (defined as "not made more than 5 mainspace edits in the last 30 days") with strong evidence of habitual LLM use, has disclosed and consented, or community consensus at a noticeboard supports it. Anything before ChatGPT's release, treated as 30 November 2022, is considered "very unlikely to be AI-generated" and stays put.

For entire pages where the AI-using editor is the sole significant contributor, a special deletion process called LLMPROD applies: if the tag "has not been removed within 5 days, an uninvolved administrator will check that the PROD complies" and the page can then be deleted. Any editor who wants to restore removed material has to take responsibility for a full review, ensuring "all content and its sources be closely reviewed and reworked as needed."

The honest caveat is that the guideline does not quantify what triggers the "very likely AI-generated" determination in the first place, how many editors have already been affected, or the false-positive rate. Presumption is a strong word, and Wikipedia's community will now be litigating what counts as sufficient evidence in noticeboard threads rather than in the content itself. The upside is straightforward: the janitorial burden on volunteer editors drops, and one of the few remaining high-trust text corpora on the open web gets a defensible way to say no to slop at scale.