A chatbot helped spot false images. The lesson did not stick.
A chatbot gave people the biggest immediate boost in judging images used to spread false claims. But on a test one week later, without help, that edge was gone. A better answer today is not proof of learning that lasts.
What the source reports
A planned experiment involved about 2,200 people in the United States, India, and Singapore. It compared ways to correct false claims made with images taken out of context. The chatbot asked questions rather than just giving answers. That approach gave the largest immediate gain in spotting misleading images. On a test one week later, people had no chatbot. The study found no lasting advantage for that group. The result is about judging image claims, not about whether chatbots improve K-12 tutoring. If the goal is learning, test what people can do later on their own.
Original source
- Title
- Chatbot Engagement Does Not Always Beget Metalearning: Evidence from Three Countries
- Author
- Jaidka, Kokil, Mujtahid, Insyirah Binte Imam, Qi, Peng, Aneja, Harshit, Mukerjee, Subhayan, Hsu, Wynne, Lee, Mong Li, Chen, Tsuhan
- Publication
- arXiv
- Date
- Saturday, September 26, 2026
Also covered by the other newsletter.