Wiki Education has concluded that editors should never copy and paste content from AI chatbots like ChatGPT directly into Wikipedia articles, following an extensive investigation into AI-generated content on the platform.
LiAnna Davis, representing Wiki Education, reports in a detailed blog post that the organization discovered a troubling pattern after analyzing thousands of articles created through its programs. The investigation reveals that more than two-thirds of AI-flagged articles failed verification, meaning the information cited did not actually exist in the sources referenced.
Wiki Education supports approximately 19 percent of all new active editors on English Wikipedia. The organization analyzed 3,078 new articles created since 2022 using an AI detection tool called Pangram. The analysis flagged 178 articles as potentially AI-generated, with none appearing before ChatGPT’s launch in late 2022.
The findings challenge common assumptions about AI-generated content. Only seven percent of the flagged articles contained citations to completely fake sources. The far more significant problem involved plausible-sounding sentences cited to real, relevant sources that simply did not contain the claimed information.
“For most of the articles Pangram flagged as written by GenAI, nearly every cited sentence in the article failed verification,” Davis writes. The cleanup process required significantly more staff time than the original article creation had likely taken.
In response to these findings, Wiki Education implemented real-time monitoring of participant edits during the second half of 2025. The organization created new training modules emphasizing where generative AI tools could help and where they should not be used. The core message remained consistent: do not copy and paste anything from AI chatbots into Wikipedia.
The intervention proved effective. While trend lines suggested 25 percent of participants might add AI content to Wikipedia articles, only five percent actually did. Staff successfully reverted all problematic content.
The investigation did identify beneficial uses for AI tools. Students reported finding generative AI helpful for identifying article gaps, locating reliable sources, checking grammar, and evaluating drafts against requirements. Crucially, 87 percent of surveyed users found AI helpful for research tasks when they critically evaluated the output.
Wiki Education supported 6,357 new editors in fall 2025, with only 217 showing multiple AI alerts. The organization processed 1,406 AI edit alerts total, though only 314 appeared in live Wikipedia articles.
Davis emphasizes that the research reveals lower AI literacy among participants than popular discourse suggests. Many participants welcomed clear guidance about appropriate AI use. The organization created supplemental training modules to address these knowledge gaps.
Researchers from Princeton University, the University of Mississippi, and Wiki Education continue studying the long-term effects of generative AI on Wikipedia contributions. Preliminary results support the organization’s findings and confirm Pangram’s accuracy in identifying AI-generated content.
Davis concludes that while AI tools can support the research process, they should never generate final text for Wikipedia. The organization remains committed to monitoring and adapting its strategies as generative AI technology evolves.
