From Dialogue to Explanation: A Framework for Analysing Engagement and Reasoning with Generative AI in Physics Learning

The rapid integration of generative artificial intelligence (AI) into science classrooms has created an urgent need for robust methodological tools to capture how learning unfolds during AI-mediated interactions. Existing re-search in AI-in-education has predominantly focused on learner perceptions or outcome-based measures, offering limited insight into the processes through which students engage with disciplinary ideas and construct scientific explanations. This article proposes a theoretically grounded methodological framework for analysing learning processes in AI-supported physics classrooms by integrating the ICAP model of cognitive engagement with the Claim–Evidence–Reasoning (CER) framework for scientific explanation. The framework operationalises ICAP for analysing student–AI dialogue and adapts CER to evaluate the epistemic structure and causal coherence of student-generated explanations produced in AI-mediated inquiry contexts. An illustrative application demonstrates how patterns of interactive, constructive, active, and passive engagement in dialogue can be systematically related to variations in the depth and integration of students’ physics reasoning. By foregrounding interactional processes and epistemic quality rather than tool-specific effectiveness, the framework supports process-oriented, theory-informed research on generative AI in science education. The proposed ana-lytic architecture is intended as a portable toolkit that can facilitate comparability across studies, support methodological transparency, and contribute to the cumulative development of knowledge about learning processes in AI-mediated physics education.