--- title: "AI for training and roleplay: practising hard conversations" description: "How AI roleplay gives people a safe, repeatable way to practise difficult workplace conversations, its limits, and where human guidance stays essential." canonical: https://peopletechrevolution.com/learn/ai/ai-training-roleplay --- [Learn](https://peopletechrevolution.com/learn) # AI for training and roleplay Roleplay prepares people for hard conversations, and it is hard to run well: actors and colleagues cost time to schedule, delivery varies between sessions, and practising in front of peers can make people self-conscious. Practice scenarios, each set in a workplace. ## The other person, replaced by a conversational AI AI roleplay replaces the other person in the room with a conversational AI. The learner speaks or types, a character built for the scenario answers in kind, and the practice can be done privately, as many times as it takes. ## Why practice in a safe space helps Handling a hard interpersonal moment, delivering difficult feedback, managing conflict, a sensitive consultation, de-escalating a customer, takes technical knowledge and emotional regulation together. Practice changes how that skill is built. A one-to-one talk with a scenario character. - Low-stakes experimentation. A learner can test a phrasing or a tone without risking a real relationship, a client's trust or their own reputation. - Repeatable practice. Unlike a scheduled workshop, the scenario is there whenever a learner needs it, so repetition can continue until confidence sets in. - Standardised scenario exposure. A large or scattered organisation can give every team member the same baseline situations. - Psychological safety. Practising privately with a virtual counterpart lowers the anxiety that makes people avoid practising at all. Your learner tests a phrasing on a colleague across the table, with nothing at risk. Your counter staff de-escalate the customer at a plain counter, as often as they need to. Your staff rehearse the hard conversation with a manager, one to one, in private. Want your team practising a real scenario like this? [Open the contact form](https://peopletechrevolution.com/learn/ai/ai-training-roleplay#contact) Ask our CEO Simon Lowe’s Digital Twin a question PTR builds the systems behind this pattern: conversational agents held to a defined purpose, approved knowledge and clear boundaries, with a path to escalate to a person. See [PTR's AI roleplay and practice](https://peopletechrevolution.com/ai/ai-roleplay-practice) for how they are put together. ## Not the same thing as VR roleplay PTR also builds VR-based roleplay, where a learner moves through a branched scenario with recorded or animated characters, evidenced in a [2024 Google CHI case study](https://peopletechrevolution.com/work/research-evidence/vr-dei-training-google-chi-2024) and two peer-reviewed [Mayo Clinic studies](https://peopletechrevolution.com/work/research-evidence/vr-nursing-leaders-mayo-clinic). A street encounter, inside a VR scenario. VR roleplay is spatial and usually pre-authored with branching choices. AI roleplay is a live, unscripted conversation with a language model. Both pair the practice with a structured debrief, the pattern supported by the research above. For the VR version, see [immersive training roleplay](https://peopletechrevolution.com/training/vr-training-and-roleplay); for an AI character inside an immersive XR scene, see [AI and XR](https://peopletechrevolution.com/learn/ai/ai-and-xr). ## What a typical session looks like Implementations vary, but an AI roleplay session usually follows the same four stages. A coach and learner review the session. - 1. Briefing and context. The learner meets the scenario, the counterpart's background and the objective, for example de-escalating an upset client. - 2. Live interaction. The learner speaks or types naturally; the character updates and answers in kind. - 3. Structured review. The system surfaces feedback against defined criteria, what worked and where tone or clarity could improve. - 4. Reflection and repetition. The learner reviews the feedback and can re-attempt the scenario straight away. ## AI roleplay against actor-based roleplay Two different resources, and a measured trial of how they compare. The same baseline scenario, run at every booth. AI roleplay Human/actor-based Availability On demand, repeatable at any time Needs a scheduled actor, coach or peer Non-verbal nuance Limited, mainly language and pacing Full body language, tone and presence Measured outcome In a 2025 randomised crossover trial, both an AI-driven and an actor-based consultation-skills session significantly improved medical students' self-rated communication skills (both p<.001); the actor-based session scored a little higher on average. [Tyrrell et al, 2025](https://formative.jmir.org/2025/1/e71667) ### A second measured result A larger trial on the same idea, run the following year. A facilitator turns over the feedback card. 162 medical students rated their own communication competence higher after a single conversation with a role-prompted GPT-4o chatbot across four clinical scenarios (mean rise 0.94 on a 0 to 10 scale, Cohen d 0.58; self-rated, not observer-scored). [JMIR Medical Education, 2026](https://mededu.jmir.org/2026/1/e88092) ## Limits, and the role human facilitation still plays AI roleplay provides the deliberate practice. Human mentors, professional judgement and peer discussion are what turn that practice into learning. Nuance and non-verbal cues. Conversational AI mainly reads language, structure and pacing; a 2025 study found removing embodiment (voice and avatar) and leaving only text measurably changed how effective participants found an AI agent for interpersonal-skills training. [Cornell/Stanford, CSCW 2025](https://dl.acm.org/doi/10.1145/3715070.3749224) Human facilitators stay necessary. They provide context, debrief the emotional reaction, check whether feedback matches organisational values, and judge overall readiness. ## Scenarios built for community and justice services A moment [community and justice teams](https://peopletechrevolution.com/work/industries/justice-community-services) need their people to handle well, and a mentor working the harder way. Mentoring that works better walking than sitting. AI companions are conversational systems built around a defined purpose, approved knowledge and clear boundaries. Conversational agents are chat and voice assistants that hold a real conversation, with configurable guardrails and a path to escalate to a person. See PTR's [AI overview](https://peopletechrevolution.com/ai) for how roleplay fits alongside digital twins and custom AI. [PTR's AI roleplay and practice](https://peopletechrevolution.com/ai/ai-roleplay-practice) [Practise a conversation with an AI now](https://labs.peopletechrevolution.com/practice) Still have a question about AI roleplay? Ask our CEO Simon Lowe’s Digital Twin a question [Open the contact form](https://peopletechrevolution.com/learn/ai/ai-training-roleplay#contact) Related PTR work and further reading [PTR's AI roleplay and practice. How the companions and their guardrails are put together.](https://peopletechrevolution.com/ai/ai-roleplay-practice) [Immersive training roleplay (VR). The spatial, pre-authored version of practice.](https://peopletechrevolution.com/training/vr-training-and-roleplay) [AI and XR. What changes when a roleplay character lives inside a headset.](https://peopletechrevolution.com/learn/ai/ai-and-xr) [What is AI? How the language model behind the character works.](https://peopletechrevolution.com/learn/ai/what-is-ai) ## Sources and further reading Studies of AI consultation and communication practice, with guidance on AI risk management. ### Web-Based AI-Driven Virtual Patient Simulator Versus Actor-Based Simulation for Teaching Consultation Skills Tyrrell et al · JMIR Formative Research · 2025 The randomised crossover trial behind the comparison table above. [Read it at formative.jmir.org](https://formative.jmir.org/2025/1/e71667) ### Training Empathetic Communication Skills in Medical Students With a Role-Prompted GPT-4o Chatbot JMIR Medical Education · 2026 The 162-student result behind the self-rated confidence figure above. [Read it at mededu.jmir.org](https://mededu.jmir.org/2026/1/e88092) ### Comparing Text-Only and Virtual Reality-Embodied Conversational AI Agents for Interpersonal Skills Training ACM CSCW Companion · 2025 The embodiment study behind the nuance and non-verbal cues note above. [Read it at dl.acm.org](https://dl.acm.org/doi/10.1145/3715070.3749224) ### AI Risk Management Framework (AI RMF 1.0) NIST · 2023 The framework behind PTR's guardrail and escalation-path language above. [Read it at nist.gov](https://www.nist.gov/itl/ai-risk-management-framework)