Why practice still belongs in the room — even with AI roleplay tools everywhere — and how to build one that actually works.
I once ran a roleplay exactly the way the facilitator guide laid it out. Instructions delivered. WIIFY sold, I thought. I asked if anyone had questions before we launched.
One hand went up. Then another. Then a third.
“Wait, I don’t get why we’re doing this.” “This doesn’t seem realistic. This doesn’t happen to me in my role.” “Couldn’t you just show us how it’s done instead?” “What would this even look like in my role?”
I hadn’t launched a roleplay. I’d launched a mutiny.
Three things had gone wrong before I ever said the word out loud. The scenario was generic. It was built for a role, not for the specific people sitting in that room, and it missed the actual gaps they were living with. The word itself worked against me. The moment people heard “roleplay,” some of them checked out before they knew what I was even asking. And underneath both of those: I hadn’t pulled from their real challenges. I’d built the scenario in advance and tried to barrel through it.
I do it differently now. Early in the session, I ask participants for their real challenges and put them somewhere everyone can see. A whiteboard, virtual or in person. When it’s time to move into practice, I point back at that list. The scenario isn’t mine anymore. It’s theirs. And I try not to call it “roleplay” if I can help it. I call it a simulation.
That dread you can feel in a room the second you say the word isn’t a roleplay problem. It’s a design problem. Roleplay has been a staple of experiential learning since the 1940s for a reason — done well, it’s one of the only training methods that puts learners in the actual skill, live, before it ever shows up in a real conversation with real consequences. Done badly, it’s theater nobody asked to be in.
You can’t fix the second problem by skipping roleplay. You fix it by building a better one.
Why We Still Build This Into the Room
Roleplay isn’t a checkbox activity — it’s the difference between learning about a skill and being able to do the skill. When a learner reads about handling a difficult conversation, that’s information. When they have to actually open their mouth, navigate pushback, and land the conversation in real time — with a safety net underneath them — that’s rehearsal. And rehearsal is what makes the first real attempt, the one with actual stakes, feel familiar instead of terrifying.
This is also where AI roleplay tools enter the conversation, because right now every L&D conference panel is debating whether they make in-room roleplay obsolete. They don’t — and the reason has nothing to do with AI’s quality as a practice partner. It has to do with when the practice happens.
AI roleplay is asynchronous by design. Learners are expected to log in on their own time, after the session, and run the scenario themselves. In my experience, that’s exactly where it falls apart. Outside of a structured moment with accountability built in — a manager checking in, a deadline, a follow-up session — that practice gets pushed to “later,” and later quietly becomes never. The intent is real. The calendar isn’t.
Live roleplay, by contrast, is built into the flow of learning. There’s no decision to make about whether to practice — the room is already practicing. That’s not a small difference; it’s the entire mechanism by which the skill sticks. Disconnect practice from the learning moment and you’ve reintroduced the exact gap that training transfer research has been warning about for decades — the same gap behind the Forgetting Curve numbers in our own Playbook.
None of this means AI roleplay doesn’t have a role. It’s a strong tool for reinforcement — extending practice into the flow of work, after the training event, where a learner can return to a scenario on their own terms. That’s a real strategy, and probably its own post. But reinforcement isn’t a substitute for the first rep. The first rep belongs in the room, where you control the conditions and the safety net is a real human, not a model.
The Tool: Building a Roleplay That Actually Works
1. Maximize practice time, minimize everything else
Borrow the spirit of the 70-20-10 framework here — popularized by Lombardo and Eichinger and a staple of L&D design ever since. The “70” is experiential, learning-by-doing time, and it’s consistently where the real skill-building happens. Apply that same ratio inside a single roleplay block: the less time you spend explaining, setting up, and demoing, the more time learners spend actually doing the skill. If you’re talking more than they are, you’ve built a lecture wearing a roleplay costume.
Practically: keep instructions tight and written down (not just spoken), keep your own modeling brief, and protect the clock for what matters — reps.
2. Build the scenario from their real challenges, not yours
A generic scenario is the fastest way to get pushback before you’ve even started. Don’t build the roleplay in advance and hope it lands. Ask participants for the real situations they’re dealing with, capture them somewhere visible, and pull the scenario straight from that list. When learners can see their own challenge on the whiteboard, “why are we doing this” stops being a question anyone needs to ask.
3. Reframe it as rehearsal, not performance — starting with the word you use
This is the single biggest lever for reducing the dread. “Performance” implies an audience judging you. “Rehearsal” implies you’re allowed to be rough, stop, restart, and get better in front of people who are rooting for you, not grading you. Say this explicitly before you start. And watch the word “roleplay” itself — for a lot of learners it’s already loaded with bad associations. “Simulation” or “practice round” can land better before you’ve said another word.
4. Give the non-participants a job
In group settings, the people watching disengage fast if all they’re doing is watching. Hand them something specific to track — a behavior to observe, a question to answer in the debrief — so the whole room is working, not just the two people in the scenario. This is the same engagement principle behind everything else on this blog: the activity isn’t the goal, engagement in service of learning is.
5. Run it twice
The first pass is almost always about content — learners are still figuring out what to say. The second pass is where the craft shows up, because the cognitive load of “what do I say” is gone and they can focus on how they say it. Don’t skip the second rep for time. It’s usually the more valuable one.
6. Debrief like it’s the main event — because it is
The roleplay isn’t where the learning happens. The debrief is. Use a real feedback structure, not “great job, who’s next.” Describe what you saw → state the impact → (for constructive feedback) offer or invite an improvement → ask for their reaction. That structure does double duty: it models the exact feedback skill you want trainers and managers using back on the job.
7. Build for the room you’re actually in
I’ve done all the setup work — instructions given, questions asked, nothing came back — and still launched breakouts straight into confusion. I jumped into one room, then another, then another. Same question in every room: “Wait, what were we supposed to do again?”
Nobody was being difficult. The information they needed had lived in my mouth for thirty seconds before they left the main room, and then it was gone. This happens in person too, not just virtually. Something about the shift from whole group to small group seems to wipe short-term memory clean. The fix isn’t better verbal instructions. It’s a visual asset — a slide, a doc, a card — that goes into the room with them, so nothing depends on what they remember hearing once.
In person, you can walk the floor and read the room in real time. Virtually, the breakout room is the roleplay room, which means the setup work has to carry even more of the weight before anyone hits “join.” Either way, the rule is the same: if it only exists in what you said out loud, it doesn’t exist once they’ve left the room.
Practice First. Reinforce Later. Don’t Skip Either.
The instinct to outsource roleplay to AI and reclaim the training-room minutes is understandable — but it solves the wrong problem. The bottleneck was never whether a practice partner is human or synthetic. It’s whether the practice happens at all, in a moment when it’s connected to the learning, instead of floating off into a calendar that never has room for it.
Build the roleplay better. Protect the practice time. Debrief like it matters. Then, if you want to extend the rehearsal into the flow of work afterward — that’s a real strategy, and a good one. Just don’t let it replace the first rep.
If you’ve built a roleplay that turned a skeptical room into an engaged one, I’d love to hear what made it work. Drop it in the comments.
Want your trainers running roleplays that build confidence instead of dread? Explore Letskillup’s Train the Trainer programs — built to help facilitators design practice that actually transfers. 🚀
The Trainer’s Blog 
Leave a comment