Developing AI-powered virtual patient chatbots for diagnostic reasoning training

Diagnostic reasoning, the cognitive process of interpreting clinical information to arrive at a diagnosis, is a critical competency in medical education, and has been taught using authentic methods such as standardized patients and peer role-playing that simulate real clinical encounters. Although effective, these approaches require substantial time, space, and cost.

To address the limitations of traditional approaches, artificial intelligence-powered virtual patient chatbots were developed between April 21-28, 2025, with a chatbot builder that incorporates natural language processing. The design was guided by 2 core cognitive models of diagnostic reasoning-the dual process model and the memory model. The chatbots simulated authentic physician-patient interactions across 3 clinical scenarios: fatigue, stomach ache, and memory loss. Students can perform differential diagnoses based on various clinical data and received immediate feedback on diagnostic accuracy.

The patient chatbots were evaluated through surveys and interviews with faculty (May 2025) and students (November 2025). Findings consistently indicated that the chatbots functioned as an effective (4.50/5) and usable tool (4.46/5) for diagnostic reasoning practice. Participants perceived that the chatbots supported "core cognitive processes of diagnostic reasoning" by prompting active hypothetico-deductive reasoning, whereas repeated exposure to different cases was viewed as facilitating pattern recognition. Faculty highlighted the chatbots' cost-efficiency and scalability within curricula, whereas students emphasized authentic "from-zero" clinical reasoning and the value of reviewing dialogue logs for reflection. Overall, the results demonstrate that the chatbots can provide a feasible and educationally meaningful environment for practicing diagnostic reasoning in medical education.

Future work should refine artificial intelligence-powered virtual patient chatbots by incorporating tiered case complexity, patient-centered language, and equity-oriented design. Effective use will require systematic instructional design and learner preparation for emotionally challenging scenarios. Larger empirical studies using both outcome and interaction data are needed to evaluate educational impact and guide integration into clinical training.