Wikipedia volunteers are checking whether sources cited by AI-written articles support the claims made about them. WikiProject AI Cleanup cofounder Ellia Lebleu describes misleading use of real sources as a persistent problem, alongside subtler signs of machine-generated prose. New checking tools can help filter submissions, but their results still need human assessment and cannot establish why detected AI content has declined.
Wikipedia volunteers face AI-written claims that cite real sources while misrepresenting what those sources contain. Ellia Lebleu, cofounder of WikiProject AI Cleanup, the volunteer effort to remove AI material that breaks the encyclopedia’s rules, described the problem in an interview with IEEE Spectrum. Even a genuine linked source can fail to support the sentence attached to it.[1]
Another volunteer has built an AI verifier to compare article claims with their cited material. Lebleu says it can help identify obvious mismatches at scale, while acknowledging that the verifier can itself be wrong. A further tool under development at the Wikimedia Foundation is Paste Check: it marks passages copied into Wikipedia from certain chatbots. English Wikipedia prohibits using language models to generate articles, and the rules appear in welcome messages and editing advice.[1]
The cleanup effort began in 2023, when repeated stock phrases and unusual conclusion sections made AI writing easier to spot. Volunteers now also examine contribution behavior, including several long articles appearing within minutes. Using Pangram, Lebleu found AI-generated material in roughly a fifth of new submissions and drafts in late 2025, with a lower share flagged in early 2026. The policy effects and the detector’s ability to keep up with changing models remain uncertain.[1]
