Beyond the prompt: structured guidance as the critical scaffold for higher-order learning with generative AI in medical education

To evaluate the efficacy of a structured pedagogical intervention in enhancing AI-assisted learning outcomes for medical students, compared to minimal guidance use of generative AI tools. A single-blind randomized controlled trial was conducted with 72 s-year medical students enrolled in an Immunology course. Participants were stratified and randomly assigned to a guided group ( n = 36), which received four weekly workshops on strategic AI prompting, critical verification, and clinical integration, or an minimal guidance group ( n = 36). Outcomes included the Immunology final exam (primary), transfer exam scores, a clinical reasoning test, validated scales for AI self-efficacy and metacognitive strategy use, and blinded analysis of AI conversation logs. The assessors of the primary and secondary outcomes were blinded to group allocation. The guided group significantly outperformed the minimal guidance group on the Immunology final exam (84.2 ± 6.8 vs. 81.8 ± 8.6, p = 0.022, d = 0.72) and the clinical reasoning test (78.5 ± 8.2 vs. 73.3 ± 10.5, p = 0.008, d = 0.65). They also reported significantly higher AI learning self-efficacy (d = 0.79) and metacognitive strategy use (d = 0.70). AI log analysis revealed substantially more sophisticated questioning, critical verification, and iterative dialogue in the guided group (composite log quality score: 7.2 ± 2.1 vs. 4.1 ± 1.8, p < 0.001, d = 1.67). Structured guidance transforms generative AI from a passive information source into an active dialectic partner in medical learning. It is superior to minimal guidance in fostering not only knowledge acquisition but also the higher-order cognitive skills and self-regulated learning competencies essential for future clinicians.