Having an argument ready and joining a roomful of people to make it are different tasks. CAiSEY puts an AI conversation between them: students practise discussing course material before the classroom discussion begins.[3]
The study estimated 0.151 fewer contributions per student in the next class after first use, a significant fall. A second use had a significant additional effect in later classes.[1]
The product is configured around an instructor’s questions, readings and cases. Students enter through a course link and receive feedback; instructors can review transcripts and summaries. The company’s research follows the later classroom contributions, making the private-to-public transition part of the question.[3][2]
The experiment followed the timetable
The 7 October preprint follows 759 MBA students in autumn 2025, each assigned AI preparation for two randomly chosen classes out of ten. Students defended positions against an opposing AI. Teaching assistants, blind to assignments, counted voluntary contributions, excluding teacher-elicited answers.[1]
After exclusions, the analysis used 504 students and 4,787 records of students in individual classes. It estimated effects of conversations lasting at least ten minutes that assignment encouraged.[1]
The later-class model estimated −0.059 contributions per student after first use and an additional 0.227 after the second, equivalent to 31% of the 0.727 sample mean. Their sum, 0.168, was not statistically significant. Neither use significantly increased participation in the class prepared for.[1]
Rehearsal has a different purpose
A student can rehearse a position, hear an objection and discover where a sentence falls apart before having to deliver it to classmates. Repeated practice could plausibly make that task easier. An alternative possibility is that an early session changes preparation in ways that initially make someone quieter. The uneven pattern leaves room for more than one explanation.
The study measured participation, not mastery, and had no human-rehearsal comparison.[1]
The uncertainty around the combined estimate matters for a teacher deciding whether to adopt this approach. A promising later increment provides a reason to investigate the sequence. It cannot, by itself, establish that prescribing two sessions improves discussion overall.
One author, Dan Wang, is also a CAiSEY co-founder.[4]
For teachers, the useful question is where a private exchange should lead. A stronger next trial would follow the ideas through the room: who contributes, which arguments classmates develop, and whether any increase in participation improves the discussion.
Sources & further reading
- How assigned AI use before class shapes active student engagement in class Wang and colleagues / arXiv · 7 Oct 2026 · Accessed 18 Oct 2026
- What happens when students discuss with CAiSEY before class CAiSEY · Accessed 18 Oct 2026
- How it works CAiSEY · Accessed 18 Oct 2026
- CAiSEY Team CAiSEY · Accessed 18 Oct 2026