From initial trust to critical reconstruction: upper secondary students’ engagement with generative AI in science learning

Introduction: This study examines how upper secondary students engage with generative artificial intelligence (GenAI) in science learning, focusing on how they move from initial trust to critical reconstruction of AI-generated content. Methods: Using a constructivist grounded theory design, the study drew on semi-structured interviews with 21 students aged 15-18 from two upper secondary schools in Southwest China. Results: The findings show a dynamic process shaped by task demands, prior knowledge, time pressure, and teacher norms. Students often began with efficiency-driven use and provisional trust in fluent, authoritative responses. They then evaluated outputs through internal consistency checking, external corroboration, and growing awareness of GenAI's limitations. Based on these judgements, students refined prompts, selectively reconstructed content, or abandoned GenAI when outputs were unreliable. Discussion: The study argues that GenAI use in science learning is best understood as a form of regulated epistemic engagement.