A new UK research project offers a genuinely better model for building psychiatric AI patients than simply adapting a general-purpose conversational system, because it incorporates medical education expertise, formal research methodology, human-computer interaction design, and, distinctively, people with lived experience of mental illness directly into the case-development process from the outset, rather than treating representativeness as something to check for after the fact.
The pilot, described
Kent and Medway Mental Health NHS Trust, Kent and Medway Medical School and the University of Kent are undertaking an eighteen-month project examining virtual patients for psychiatric history taking, communication, empathy and clinical reasoning. The scenarios have been co-produced with people who have lived experience of mental illness, with a student pilot phase and a subsequent trial planned. This structure, formal academic and clinical partnership plus direct lived-experience involvement in scenario development, is a genuinely different starting point from a commercial platform building psychiatric cases through its general case-writing process.
Why psychiatry is particularly suitable for conversational simulation
Psychiatric assessment is, in a specific sense, unusually well matched to conversational AI's actual strengths, since the core assessment tools, history taking, exploring thought content, assessing risk through dialogue, are conversational by nature in a way that a cardiovascular or respiratory examination is not, removing the physical-examination limitation this cluster's PACES coverage identifies as the sharpest gap elsewhere in this category.
Why it is also particularly risky
The same conversational centrality that makes psychiatry well suited to simulation also raises the stakes of getting the simulation wrong, because psychiatric presentations carry a genuine risk of stereotyped or clinically inaccurate portrayal doing real harm to how future clinicians understand and relate to patients experiencing these conditions, a risk this cluster's dedicated CASC-preparation analysis raises directly regarding stereotyped portrayals of psychosis, personality disorder and substance use specifically.
The importance of lived-experience co-production
Involving people who have actually experienced the conditions being simulated in the case-development process addresses this risk directly, in a way clinical expertise alone cannot fully substitute for, since accurate clinical knowledge about a condition and authentic understanding of how that condition is actually experienced and communicated are genuinely different kinds of expertise. This is the specific design feature this article holds up as a benchmark the wider commercial market should be measured against, and the specific question worth asking any commercial psychiatric-simulation product directly: was lived experience involved in developing this content, and if so, how.
Design considerations the pilot's structure surfaces
Avoiding stereotyped portrayals of psychosis, personality disorder and substance use, the specific risk lived-experience co-production is designed to address. Simulating guardedness, ambivalence and inconsistent accounts authentically, since real psychiatric presentations frequently involve exactly these features, and a simulated patient that is always straightforwardly forthcoming teaches candidates against a version of practice that does not reflect real encounters. Risk-assessment language, requiring particular care in how a simulated patient discusses risk and how a candidate's response is assessed. Capacity and safeguarding, specific content areas demanding the same careful, accurate treatment any genuine clinical encounter requires. The emotional impact on learners themselves, a genuine consideration in psychiatric simulation design that deserves explicit attention rather than being assumed negligible because the patient is not real. Whether and how AI should simulate distressing content within a training context at all, a genuine design and ethical question any programme building this kind of simulation needs to work through deliberately, with appropriate boundaries, clinical oversight and pedagogical purpose guiding the decision rather than technical capability alone. And human debrief and escalation support for learners after emotionally demanding scenarios, ensuring that simulated difficult encounters are followed by the same kind of support a real difficult clinical encounter would warrant.
Outcomes worth measuring
The trial's planned evaluation offers a template for what rigorous evidence in this specific category should look like: knowledge, communication skill, empathy, stigma toward the conditions simulated, candidate confidence, performance with an actual human patient afterward, and, critically, unintended harm, checking explicitly that the simulation has not produced negative effects on how candidates understand or relate to patients with these conditions, a possibility formal evaluation should test for directly rather than assume absent.
What commercial products can learn from this
Placing this research pilot's co-production approach alongside commercial products in this category, Simsbuddy's, CASC Master's and PassMRCPsych's psychiatric content among them, the useful question is not which currently has more case volume, it is which has engaged with the specific representational and clinical-accuracy risks this pilot's structure was built to address, and whether any commercial platform's psychiatric case-development process involves anything comparable to formal lived-experience co-production, a standard this article argues the whole category should be measured against rather than treating case volume or voice quality as the primary differentiator in this specific, higher-stakes examination category.
Frequently asked questions
When will results from the Kent and Medway pilot be available?
The project is planned across an eighteen-month timeline including a student pilot phase and a subsequent trial, with formal outcome evidence expected as that trial phase completes and reports.
Should commercial psychiatric-simulation platforms be held to the same evidence standard as an NHS research pilot?
Not necessarily the same formal trial methodology, though the underlying principle, lived-experience involvement in case development and honest evaluation of whether simulation produces the intended learning without unintended harm, is a reasonable standard to expect from any product used to prepare clinicians for real psychiatric encounters.
How does this pilot relate to the wider evidence on AI virtual patients improving OSCE performance?
It sits alongside this cluster's broader evidence review of that question, adding a psychiatry-specific and co-production-focused dimension that the existing 2026 quasi-experimental study, which did not focus specifically on psychiatric content, did not directly address.