Student-Initiated Questioning in LLM-based Physics Tutoring: Metacognitive Regulation and Dialogic Uptake

Student questioning is fundamental to physics learning, yet traditional intelligent tutoring systems often constrain this inquiry by following rigid instructional protocols. This study investigates how a Conversational Intelligent Tutoring Systems (CITS) based on a Large Language Model (LLM) transforms student engagement in thermodynamics education. Eighteen undergraduate students interacted with the CITS, generating 112 spontaneous questions that were analysed using a dual framework of metacognitive regulation and dialogic uptake. Our qualitative analysis characterises how students moved between passive content consumption and active knowledge construction through their own questioning. The findings show that students’ questions were not merely informational but were distinct acts of metacognitive regulation (planning, monitoring and evaluation). Furthermore, the vast majority of these student-initiated dialogues were of high quality, demonstrating deep cognitive engagement (Level 2 and 3 uptake). Critically, sustained Monitoring, in which students continuously tracked and tested their understanding during problem-solving, was significantly associated with high-level dialogic interaction and accounted for most instances in which students extended concepts to new contexts (Applying and Connecting). These findings suggest LLM-based systems can restore learner agency in physics education by creating conversational affordances that support the spontaneous, critical inquiry essential for deep conceptual understanding.