The challenge of teaching patient-centred communication in physiotherapy is not new. Traditional simulation methods—peer role-play or actor-based scenarios—have their merits but also clear limitations. Role-play often lacks authenticity, and actor-led simulations, while valuable, are difficult to scale. What’s become increasingly clear is the need for accessible, flexible approaches that still maintain the depth of real-world practice. That’s where AI has started to offer something meaningful.
This work began with a simple idea: could generative AI simulate patient interactions in a way that felt realistic enough to support communication skill development?
Using Claude and Blackboard Ultra’s AI Conversations digital tool, we developed a series of simulated patient personas grounded in clinical scenarios. Prompts were deliberately detailed to encourage consistent character responses. Students engaged in 20-minute text-based conversations designed to replicate subjective assessments—some with paediatric patients, others with adults managing long-term conditions or cognitive decline.
These interactions were not standalone activities. They were integrated within a broader learning design grounded in simulation pedagogy—specifically, the principles of fidelity, psychological safety, and structured debriefing. The conversations were followed by reflective discussions and formative feedback, using submitted transcripts as learning artifacts.
Feedback from students was consistently positive. Most appreciated the low-stakes environment, the ability to work at their own pace, and the realism of the AI responses. This was especially valuable for international learners and those with varying levels of clinical experience, who often benefit from more time to process and practice communication techniques.
There were challenges of course, Claude occasionally broke character, and its limitation to English only excluded opportunities for multilingual engagement. Still, the activity proved fairly effective at scale. Seemingly unlimited numbers of students could take part at once, all engaging with different patient personas in parallel, however, the ‘bandwidth’ on Claude’s servers significantly reduced the capacity for questions. Responses were sometimes limited to as many as one/two per student – catastrophic for the activity. Blackboard Ultra’s AI Conversations tool completely negates these challenges however, with unlimited responses, participants and the opportunity to provide direct audio/video/text feedback. English only remains a limitation however.
From a practical and logistical teaching perspective, it reduced dependency on service users or simulated patients while maintaining an element of emotional and cognitive complexity.
What emerged was a model of AI-supported learning that complements, rather than replaces, existing educational approaches. The simulation isn’t perfect—but it doesn’t need to be. It’s structured enough to prompt genuine interaction and flexible enough to support personalised feedback. More importantly, it allows us to embed communication practice in a scalable, inclusive way.
As AI tools become more prevalent in education, the focus must remain on pedagogy first. The technology is only as effective as the thinking behind it. In this case, simulation pedagogy provided the structure; AI simply extended its reach.
Dean Walker 