Human-GenAI collaboration in clinical reasoning education: a systematic review of roles, risks, and curricular implications

Background Generative AI (GenAI) is increasingly being explored as an interactive educational tool in medical education. Although emerging studies describe benefits for clinical reasoning, the pedagogical roles, conditions of use, and cognitive risks of human-GenAI interaction remain fragmented. Drawing on clinical reasoning and cognitive load perspectives, this review synthesized empirical evidence on how GenAI roles align with reasoning phases and how reported educational outcomes coexist with interaction risks. Methods A systematic review was conducted following a search of PubMed , Embase , Web of Science , PsycINFO , and ERIC (November 2022–April 2026). Three reviewers screened titles and abstracts, and two reviewers independently assessed each full-text publication, yielding 28 empirical studies. Data were extracted and synthesized using a deductive-inductive approach guided by Best-Fit Framework Synthesis, with dual coding against established taxonomies of GenAI pedagogical roles, clinical reasoning phases, and interaction outcomes and risks. Results The included studies positioned GenAI in four overlapping roles: Simulated Patient, Socratic Tutor, Clinical Co-reasoner, and Evaluator. These roles were used across information gathering, problem representation, differential diagnosis, and management decision-making. Many studies reported performance gains or positive learner perceptions, but the outcome measures were heterogeneous and mostly assessed short-term effects. Higher-rigor and mixed-methods studies also identified automation bias, inappropriate reliance, prompting burden, and increased confidence despite incorrect AI-supported decisions. Conclusion Current evidence suggests that GenAI may support selected phases of clinical reasoning when its role is aligned with the learner’s level and the instructional task. However, the diversity and short duration of the available studies preclude universal curricular prescriptions. AI clinical literacy, including verification and appropriate trust calibration, should therefore be considered an evidence-informed curricular priority that requires longitudinal evaluation.