Supporting CS1 Students in the Age of Generative AI: Behavior-Informed Interventions to Promote Self-Regulated Learning

• Kacey Dove, Computer Science, University of Delaware

• Austin Cory Bart, Department of Computer and Information Sciences, University of Delaware

Introductory computer science (CS1) courses have experienced disturbingly high fail rates recently. A possible explanation is students overrelying on Generative AI tools like ChatGPT while completing their programming homework, leading to poor performance on proctored exams, which now make up the majority of students’ final grades. Instructors seek to help, but need to be able to distinguish between students who are struggling because of overusing AI, those with low motivation or engagement in the course, and those who simply have poor study habits. To better understand these categories of students, we analyze student behavioral data from CS1 courses before and after the release of GPT-5 using five metrics: exam performance, homework performance, time spent on homework, sudden large code change events, and suspicious code characters. The last three metrics are combined into an AI-use indicator that we use, along with students’ exam and homework performance, to identify behavior-based target groups. The comparisons between pre- and post-GPT-5 semesters show an increase in students with higher AI-use indicators and lower exam performance. From these results, we propose the following targeted interventions to promote self-regulated learning principles: periodic automated feedback messages; required TA office hours visits early in the semester guided by identified issues; lectures and assignments on responsible AI use and self-regulated learning skills; and post-exam live coding reflections. These interventions are designed to benefit all CS1 students while providing targeted support for those displaying behaviors associated with AI overreliance, low motivation, or academic struggle.