Better homework did not mean better exam scores
A study of 153 college students found that earlier AI access improved math homework scores, but not exam scores in the study’s comparison period. More chatbot messages later went with lower exam scores; that link alone cannot show the chatbot caused the drop.
What the source reports
Chen and colleagues studied 153 students in an undergraduate course about math proofs. Students got access to an AI chatbot and homework feedback at different points in the course. During the early comparison period, the group with access scored 2.56 points higher on homework. The groups did not differ significantly on the later exam. After everyone had access, heavier chatbot use was linked to lower later exam scores. That link does not prove the chatbot caused lower scores. The feedback tool asked students to submit their own proof before it responded; its use was not linked to lower exam scores. This is one college course, not evidence about K-12 students.
Original source
- Title
- Supporting or Replacing Students' Reasoning? Cognitive Offloading in AI-Assisted Proof Learning
- Author
- Eason Chen, Sophia Judicke, Kayla Beigh, Xinyi Tang, Isabel Wang, Nina Yuan, Zimo Xiao, Chuangji Li, Shizhuo Li, Reed Luttmer, Shreya Singh, Maria Yampolsky, Naman Parikh, Yvonne Zhao, Meiyi Chen, Scarlett Huang, Anishka Mohanty, Gregory Johnson, John Mackey, Jionghao Lin, Ken Koedinger
- Publication
- arXiv
- Date
- Friday, October 23, 2026