Metacognitive Scaffolding in the Age of GenAI: A Behavioral Analysis of Student–Chatbot Interactions During Course Selection

Course selection presents a persistent challenge for students who often have difficulty articulating clear goals, integrating multiple considerations, and aligning academic choices with personal and professional aspirations. This study investigates whether concept mapping, as a metacognitive scaffolding tool, may shape how students interact with Generative AI (GenAI) systems during academic decision-making. In a randomized controlled experiment, 180 undergraduates at a polytechnic university in China were assigned to either a GenAI-only condition or a GenAI + Concept Map condition. After excluding 3 outlier participants, 177 students were included in the final analysis. Controlling for prior academic performance via ANCOVA, students with concept-map support showed different interaction patterns: they had a longer maximum consecutive-question chain within a session (GPA-adjusted means: 11.92 vs. 9.07 questions), formulated longer questions (15.27 vs. 11.93 words), and spent more time per conversation session on average (8.05 vs. 6.77 min). An analysis of conversation content showed that the concept-map group discussed a wider range of course selection factors (covering 4.46 vs. 3.66 main dimensions and 8.70 vs. 6.36 detailed factors). Epistemic Network Analysis further suggested that concept-map users linked different factors more frequently in their conversations, connecting academic requirements with career development, intrinsic interests, and external recognition in their discourse. Notably, these group differences remained after controlling for GPA in the ANCOVA models. These findings suggest that metacognitive scaffolding may reshape the way students engage with GenAI, with concept-map users shifting from brief exchanges to extended conversations covering multiple integrated factors related to their academic choices.