Wikipedia implements definitive ban on AI-generated article text
Wikipedia leadership cites hallucinations and fake citations as reasons for the 2026 crackdown
Wikipedia has implemented a sweeping new policy officially banning the use of large language models (LLMs) to generate or rewrite article text, effective Friday.
The decision, ratified by a decisive community vote of 40 to 2, reflects an urgent effort to protect the encyclopedia's core principles of neutrality and verifiability.
The updated guidelines replace previous, more ambiguous rules that only prohibited creating "new articles from scratch," closing a loophole that editors had used to update existing entries with synthetic text.
Wikipedia founder Jimmy Wales previously described the current state of AI-generated information as a "mess," noting that the models are still "nowhere near good enough" for encyclopedic standards.
The new mandate allows for only two narrow exceptions: translation and basic copy-editing. Editors may use AI to suggest stylistic refinements to their own human-written prose or to produce first-draft translations from other language editions of Wikipedia. However, these tools are strictly prohibited from introducing any "content of their own."
Furthermore, the policy requires that any editor using AI for translation must be proficient in both languages to catch the "hallucinations"—plausible-sounding but false information—that LLMs frequently produce.
This "Human Review Mandate" comes at a critical time for the Wikimedia Foundation. In late 2025, the organisation reported an 8 per cent drop in human page views, which it attributed to the rise of AI-powered search summaries that provide answers directly to users without linking back to the source.
While the Foundation has signed commercial deals with tech giants like Microsoft and Meta to allow paid data scraping, the volunteer community remains wary of "AI slop" degrading the site’s reputation.
To combat this, the "WikiProject AI Cleanup" has been established to identify and delete the estimated 5 per cent of new articles that have been flagged as potentially machine-written since the AI boom began.