Medical students’ and clinical teachers’ experiences and perceptions of using AI simulated patients in psychiatry clinical skills training: a qualitative analysis

Background

Simulation-based teaching is central to undergraduate psychiatry education, supporting the development of key clinical skills. Recent advances in generative artificial intelligence (AI), and digital avatars, have enabled the development of AI-driven simulated patients (ASP) that may offer flexible, repeatable learning opportunities. However, there is limited evidence regarding the perceptions of medical students regarding the use of such tools in psychiatry education.

Aims

To explore medical student’s attitudes and beliefs towards the use of AI simulated patients for psychiatry clinical skills training, to inform future educational development and integration.

Method

Medical students were invited to participate through university mailing lists and departmental and student networks. Interested participants self-enrolled by responding to the invitation. Clinical teachers were recruited through the Royal College of Psychiatrists’ South East Division newsletter. All participants gave both written and verbal consent. A total of sixteen medical students participated across two focus groups. Nine clinical teachers participated across three groups. Each session included a 10-minute hands-on interaction with an exemplar AI based simulation tool, which was developed by the research team. The tool was built in May 2025 using OpenAI’s ChatGPT voice feature (Advanced Voice Mode, powered by GPT-4o), Streamlit (v1.45), and LangChain (v0.3.x). Each AI simulated patients’ interaction was followed by a facilitated group discussion. Sessions were audio-recorded and transcribed verbatim for subsequent thematic analysis.

Results

Participants identified both potential benefits and limitations of AI simulated patients in psychiatry education. It was seen as useful for early communication skills training, by offering flexible, low-pressure practice. However, concerns were raised about how well the avatars functioned technically and how accurately they could present people experiencing mental illness, as well as their limited realism compared with actor-based teaching and real patient encounters.

Key priorities for future development included improving emotional realism, expanding scenario options, and ensuring closer alignment with communication skills teaching and institutional systems to support meaningful curriculum integration.

Conclusions

Participants recognised the potential of AI simulated patients for undergraduate psychiatry education but identified key barriers to engagement, particularly system reliability and trust, realism representation and institutional access. Addressing these considerations will be important for effective integration, supported by ongoing student involvement.