Abstract
<title>Abstract</title> <p>Clinical reasoning underpins clinical expertise, yet fostering its development remains challenging. Effective clinical reasoning requires students to monitor and control how their reasoning develops as a clinical problem unfolds. These processes are commonly described as metacognition. Existing approaches provide limited timely and adaptive support for students’ metacognition as reasoning evolves. Generative artificial intelligence (GenAI) creates opportunities for delivering such support, but how GenAI-mediated metacognitive scaffolding should be designed remains unclear. Using design-based research, we designed and iteratively refined such scaffolding within a simulated history-taking environment across three cycles. Fixed predefined questions functioned mainly as task reminders. Socratic questioning provided broader metacognitive support but was imbalanced and was sometimes experienced as endless and as knowledge testing. Adaptive on-demand scaffolding provided more balanced support and enabled students to select support aligned with their evolving reasoning needs. Effective GenAI-mediated metacognitive scaffolding requires context integration, balanced support, and preservation of student agency.</p>