Atypical Interaction Conference

At the Atypical Interaction Conference 2025, our team presented work from our ongoing collaboration between conversation analysis researchers and AI engineers. Together, we have been rethinking how Dora — a conversational AI agent used in telephone-based clinical consultations — engages with patients, and whether it can do so with genuine relational sensitivity.

The project begins on the phone. We analysed real telephone consultations between human clinicians and patients in a UK fracture liaison service, examining the conversational moves that make clinical interactions not just informative, but supportive. How do skilled clinicians build rapport? How do they display empathy when a patient describes pain? What makes a patient feel genuinely heard — and more willing to engage with their care?

Our preliminary findings reveal that empathy in this setting is often expressed through subtle affiliative work — for instance, an embedded assessment woven into a response to a patient's disclosure about their symptoms. These are small, precise conversational gestures that carry significant relational weight.

Armed with these insights, we then redesigned the conversational framework prompts guiding Dora's behaviour, embedding these practices directly into the system. The final stage: analysing clinical trial calls between Dora and real patients to ask how well these design changes translate — and whether patients respond to AI-delivered empathy in the same way they do to human clinicians.