Self-regulated approaches to students’ generative AI use in higher education

This study examines how undergraduate students integrate generative AI tools into home exams and how these practices relate to self-regulated learning (SRL) and academic performance. Using a mixed-methods design, we analyzed exam responses, self-reports, and performance data from 891 students enrolled in BA-level qualitative methodology courses (2023–2024). Students were permitted to use AI if they disclosed their usage. Three modes of AI engagement were identified: instrumental (task completion with minimal evaluation), ideational (idea generation and selective integration), and reflective (critical evaluation and feedback-driven refinement). At a general level, AI use was associated with increased efficiency and somewhat higher performance than non-use. However, these aggregate patterns masked substantial variation across modes of use. Instrumental use was associated with weaker performance in some tasks, including definition tasks where reliance on generic AI-generated responses did not align with course-specific criteria. In contrast, reflective engagement with AI was associated with stronger and more consistent performance. Interpreted through an SRL framework, these findings suggest that outcomes of AI use depend on how students regulate its use. Rather than uniformly supporting or undermining learning, AI appears to amplify differences in students’ self-regulatory capacities. This highlights the importance of supporting students in developing skills for critical and reflective engagement with AI.