Designing a source-grounded Socratic anatomy coach using artificial intelligence to support anatomical reasoning during clinical clerkships

Anatomical knowledge must remain retrievable and clinically usable during clerkships, yet previously learned anatomy is often least accessible when students need it most. Generative AI is increasingly available at the point of need, but unstructured use risks displacing the cognitive work required for durable learning. The educational problem is therefore not whether AI can answer anatomy questions, but how a chatbot can be designed so that it supports learning without replacing anatomical reasoning. In this discursive article, we present the design of a source-grounded Anatomy-AI Coach for medical students during clinical clerkships at Radboudumc. The current proof-of-principle covers musculoskeletal and neuroanatomical questions and is intended for preparatory, just-in-time, and reflective use around patient encounters. The design is organized around a five-phase interaction protocol: goal clarification, concise answer with source citation and micro-action, multimodal bridge, Socratic coaching with a mirror step, and reflective closure. The protocol draws on self-regulated learning, the ICAP framework, and emerging work on cognitive offloading and AI as a cognitive mirror. We describe guardrails, scope restriction, source grounding, uncertainty communication, clinical boundaries, and privacy as didactic design features rather than purely technical constraints. This article is part of a larger design-based research project; future phases will evaluate factual accuracy, pedagogical uptake, and effects on anatomical reasoning during clerkships. The present contribution offers a design blueprint rather than outcome data, specifying how the interactional architecture of an anatomy chatbot can be aligned with workplace learning.