Study finds banning ChatGPT in class hurts students

Thibault Schrepel, a professor at Vrije Universiteit Amsterdam, tracked student performance over two years to analyze the impact of ChatGPT in the classroom. The research involved 66 students in 2024 and 164 participants in 2025. Students in a "Law of AI" course were divided into three groups to revise a provision of the EU AI Act within 20 minutes. The first group was banned from using AI, the second used ChatGPT without guidance, and the third received structured training in legal prompt engineering.

The results challenged traditional academic assumptions, as the group banned from using AI finished last in both years. Members of the AI-free group suffered from what Schrepel termed "idea exhaustion" after 10 to 15 minutes, while the unguided ChatGPT group performed surprisingly well on exams despite initially accepting flawed AI suggestions. While the trained group excelled in 2024, the performance gap between the groups narrowed in 2025 because students had already developed general familiarity with chatbots in their daily lives.

These findings contrast with policies at institutions like UC Berkeley Law, which has banned AI from most graded work. However, other data highlights the risks of unsupervised AI use. A UC Berkeley study of over 500,000 grades showed a 13 percentage point jump in top grades for writing and programming courses after ChatGPT debuted, though proctored exams showed no change. Similarly, a study of more than 26,000 K-12 students in China found that while AI improved homework marks, exam scores dropped by up to 24 percent when AI replaced independent critical thinking.

For educational practitioners and university administrators, this research indicates that blanket bans are counterproductive. Instead of blocking AI, instructors should design curriculum-integrated training that teaches students how to critically evaluate AI outputs. Schrepel suggests restructuring traditional assignments, such as master's theses, to focus on empirical or practice-oriented work where the thoughtful application of AI is actively evaluated and graded.