Wikipedia Bans AI-Generated Text in Articles Under New Editing Policy

Wikipedia editors have implemented a new policy prohibiting the use of large language models (LLMs) to generate or rewrite article content, citing concerns over accuracy, hallucinations, and fabricated sources. This move reflects growing apprehension within the community about AI-generated text conflicting with the platform’s core content policies, particularly regarding verifiability and reliable sourcing.

Policy Details and Exceptions

The updated guideline explicitly states that text generated by LLMs often violates Wikipedia's content standards. While the generation or rewriting of articles using LLMs is banned, the policy allows for limited AI assistance. This includes basic copy edits suggested by AI tools for an editor's own writing, provided no new information is introduced. Editors are strongly advised to meticulously review all AI-generated suggestions.

Although the policy does not specify immediate penalties for AI content misuse, repeated violations can be considered a “pattern of disruptive editing” under Wikipedia’s disclosure guidelines, potentially leading to account blocks or bans. Users have recourse through an appeal process to reinstate their accounts.

Expert Perspectives on AI in Editing

Linguistics professor Emily M. Bender from the University of Washington noted that while some AI uses, like advanced spell or grammar checkers, might be reasonable, distinguishing between editing and content generation can be challenging. Bender highlighted that LLMs lack the accountability inherent in human contributions, as they do not operate based on belief or personal responsibility. She warned that widespread use of AI-generated edits could degrade Wikipedia's overall value and reputation.

Joseph Reagle, an associate professor studying Wikipedia's culture, echoed these concerns, stating that the community is wary of AI-generated prose due to its limitations in accurately characterizing reliable sources, often resulting in “hallucinated” claims and fabricated sources. He also pointed out the irony that many LLMs are trained on Wikipedia content, while human visits to the site have declined as search engines and chatbots increasingly provide direct answers.

Context and Future Implications

The Wikimedia Foundation has previously established agreements with major AI companies like Microsoft, Google, Amazon, and Meta to license Wikipedia content through its Enterprise product. Reagle observed that while content reuse is permitted by licenses, there is some resistance among Wikipedians to services that appropriate community content, leading to a potential influx of AI-generated