Ellia Lebleu, a founding member of the Wiki Project AI Cleanup, leads volunteer efforts launched in 2023 to identify and remove AI-generated content from Wikipedia. The project began after editors noticed articles repeating distinctive phrases like "stands as a testament" and "rich cultural heritage," traits typical of ChatGPT output. Today, volunteers develop tools to detect more subtle structural patterns, including the "rule of three" and negative parallelisms that accumulate across articles.
Lebleu, pursuing a Ph. D. in AI at the Inria Center of the University of Rennes, explains that suspicious activity includes authors producing thousands of words in minutes. Beyond text, fake or misleading citations remain a major concern, with AI sometimes hallucinating sources or misrepresenting real ones.
In late 2025, AI-detection tools flagged approximately 20 percent of new articles and drafts, decreasing to 15 or 16 percent in early 2026. Wikipedia has prohibited editors from using LLMs to generate articles, and tools like an AI source verifier and Paste Check aim to filter obvious cases at scale. However, open-weight models can bypass commercial restrictions.
It remains early to determine if policies are effective or if models are simply improving faster than detection methods.
Source: Hacker News · Summarized by HeadlinesBriefing