When the AI model was instructed to persuade users without using facts or evidence, its debunking effect on conspiracy beliefs was completely eliminated.
Notes on verification
Corroborated by Science/AAAS news coverage, MIT Sloan reporting, and two independent academic papers citing the same follow-up experiment, all converging on the finding that removing factual/evidentiary content eliminated the AI's debunking effect on conspiracy beliefs.
Sources
- Chatbots are surprisingly effective at debunking conspiracy theories (technologyreview.com)
- https://www.science.org/content/article/ai-chatbot-shows-promise-talking-people-out-conspiracy-theories (science.org)
- https://mitsloan.mit.edu/ideas-made-to-matter/mit-study-ai-chatbot-can-reduce-belief-conspiracy-theories (mitsloan.mit.edu)
- https://arxiv.org/pdf/2510.01537 (arxiv.org)
- https://arxiv.org/pdf/2506.09292 (arxiv.org)