For most flagged articles, nearly every cited sentence failed verification (wikiedu.org)

Wiki Education has examined the impact of generative AI on Wikipedia editing, revealing significant challenges associated with AI-generated content. Their research focused on participants in their programs who utilized tools like ChatGPT, finding that a staggering majority of flagged articles contained unverifiable or hallucinated information. The organization emphasized that editors should avoid copying and pasting AI outputs directly into Wikipedia, as this compromises the integrity of the encyclopedia. Despite some early concerns about generative AI usage, only 5% of participants ended up incorporating AI-generated text into live articles, which Wiki Education successfully reverted. The organization has taken proactive measures by introducing a training module for new editors that advises on appropriate uses of generative AI, such as identifying article gaps and finding reliable sources, rather than drafting text. They employed Pangram’s AI detection system to monitor edits in real-time, successfully reducing the prevalence of problematic AI-generated content. This initiative not only preserves Wikipedia’s commitment to verifiability but also highlights the potential for AI to assist in the editorial process when used appropriately. Overall, the findings underscore the importance of critical evaluation and human oversight in content creation within the Wikipedia environment.