Classroom simulation links chatbot style to student dependence
This is not a real classroom study, and its numbers are not student outcomes. It still shows why a bot's advice style can become a design choice. In repeated help sessions, the wording of support may steer whether simulated students keep asking the AI.
What the source reports
Rin Tamai and Yuya Dan built a fake classroom. It used 20 virtual students run by software. Students could talk, argue, ask friends for help, or ask a counselor AI. The counselor used one of six styles: affirming, listening, solution-focused, reality-redirecting, inciting, or blaming. A second model updated stress, happiness, self-reliance, sociability, and AI dependence. In a logged 50-day rerun, there were 153 AI consults and no failed API calls. API means a way for one program to talk to another. Affirming and inciting styles raised AI dependence to 0.55 and 0.63. Control was 0.11. The authors say these are simulation values from single runs, not measured effects on real students.
Original source
- Title
- How a Chatbot's Response Style Shapes a Classroom: A Multi-Agent Simulation of Students Consulting AI
- Author
- Rin Tamai, Yuya Dan
- Publication
- arXiv
- Date
- Friday, September 4, 2026