AI roleplay and practice
Not every difficult conversation needs a headset. PTR also builds AI companions and conversational agents for practice on a laptop or phone: domain-grounded conversational systems designed around a defined purpose, approved knowledge and clear boundaries, that hold real conversations with configurable guardrails and escalation paths.
AI roleplay lets someone practise a difficult conversation on a laptop or phone, as many times as it takes, before trying it with a real person, with the same briefing, structured review and human-escalation pattern PTR uses in its VR practice work.
How it works
A typical AI roleplay session runs in four stages: briefing and context (the learner is introduced to the scenario, the counterpart and the objective), live interaction (the learner speaks or types naturally and the AI persona responds), structured review (feedback against defined conversational criteria, highlighting what worked and what to adjust), and reflection and repetition (the learner tries the same scenario again with what they have just learned). Practising privately, before a real conversation with a real person, removes the anxiety of trying something new in front of colleagues, and the scenario can be replayed as many times as it takes to build confidence.
None of this replaces human mentorship or peer discussion. Conversational AI mainly reads language, structure and pacing, so it can misjudge intent in an open-ended exchange if a scenario's guardrails are too loose, and it cannot fully interpret body language or cultural nuance the way a human facilitator can. PTR builds AI roleplay on top of human oversight, retrieval over an organisation's own approved knowledge so scenarios reflect real policy and practice, and clear escalation paths back to a person when a conversation needs one.
Featured work
PTR's most established evidence for the practice-and-debrief pattern behind this capability comes from its VR work, where the same briefing, live interaction, structured review and repetition structure has been independently evaluated:
of QA testers (10 of 11 who responded) said the practice scenarios felt real to them, in the published Google CHI 2024 case study of PTR's VR practice-and-debrief format.
average trainee satisfaction, self-reported by 7 of the 26 trainees who took part and returned the post-training survey. Small, self-selected sample; read the full case study for its own stated limitations.
These numbers describe PTR's VR practice format specifically, not an AI-specific study; they are the closest documented evidence for the shared practice-and-debrief pattern this AI capability is built on. See also the Mayo Clinic case study for a second, larger-scale example of the same approach, and Immersive Training & Roleplay for the VR version of this page.
Try it
Read AI for Training and Roleplay for the full picture of how a typical session runs, what it is good at, and where it falls short of human facilitation and peer discussion. See AI practice running as a real build at LABS DEMOS, or explore the wider platform, including how a practice session connects back to a team's own tools, at AI STUDIO DEMOS.
Sources and further reading
- NIST AI Risk Management Framework - National Institute of Standards and Technology, 2023
Immersive training and roleplay · Digital twins · Custom AI experiences · AI for training and roleplay