Realistic practice works, but only if learners feel safe
Sarah Mateljan
Too much realism makes learners shut down. How to design role-play practice that challenges people without making them feel threatened.
When practice feels threatening, learners stop learning and start protecting themselves. The most effective practice isn’t the most realistic. It’s realistic enough to challenge people, safe enough that they stay with it, and pitched to their experience level. Finding that balance takes testing, not guesswork.
Our co-founder and CEO, Sarah Mateljan, saw this play out at the WA Museum Boola Bardip with her five-year-old son. He was fascinated by the life-size whale shark model hanging from the ceiling. He was also so unsettled by how real it looked that he refused to walk underneath it. They went the long way round.
The exhibit did its job: it showed exactly how big a whale shark is. But it was realistic enough to feel threatening, so he stopped engaging with it. Workplace training works the same way. Here’s how to design practice that’s realistic enough to stretch people, and safe enough that they stay with it.
Why realism matters
People don’t build skills by reading about them. They build skills by trying, getting it wrong and trying again, ideally in situations that feel like the real thing.
Realism is what makes practice transfer to the job. A student who has rehearsed talking with a distressed client is better prepared for placement than one who has only read the case notes. A manager who has practised a performance conversation is more likely to stay calm when the real one gets difficult.
Role-play has always been one of the best ways to do this. The problem has been scale: trainers’ time is limited, and peers watching can make it more stressful than useful. AI role-play solves the scale problem. That makes the question of how realistic to make it more important, not less.
When realism tips into threat: the story of Mr Rude
Video Arts, a long-standing provider of workplace training content, built a customer service practice conversation around a character called Mr Rude. As the name suggests, he was designed to be critical and aggressive, the kind of customer every frontline worker eventually meets.
It was realistic. It was also too much. Learners found Mr Rude so confronting that they focused on getting through the conversation rather than learning from it.
So the Video Arts team toned him down. He’s still difficult, and he still pushes back, but now he’s a challenging and useful practice partner rather than an ordeal. Like the whale shark, the original version was accurate. It just crossed the line from stretching learners to threatening them.
Getting that level right isn’t something you can calculate in advance. It takes testing, trial and error, and listening to how learners actually respond.
Pressure stops people experimenting
Feeling threatened isn’t only about aggressive characters. High stakes have the same effect.
Sarah found this when she first tried building AI agents for business admin. Those agents were going to be part of her daily work, so they had to be right, and after hours of frustration she wasn’t learning much. She switched to a low-stakes project, an agent that checks supermarket specials, and learned more in one go than in all the hours before. With nothing riding on it, she was free to experiment.
Learners are no different. If every attempt feels like it counts, they play it safe, stick to what they already know and avoid the mistakes that teach them the most. Safe practice gives them permission to try something new.
How to design practice that’s realistic and safe
The goal isn’t hyper-realism. It’s enough realism that people are challenged, but not so much that they feel threatened. These five steps apply whatever tool you use, AI or human role-play.
Start with your learners’ experience level. Junior learners building confidence need a gentler starting point than experienced staff sharpening a skill. Set the difficulty for the people in the room, not for the hardest version of the situation.
Keep the situation real, and adjust the intensity. The scenario, the language and the stakes for the character should feel true to the job. What you adjust is how hard the character pushes back.
Make it private. Anonymous practice with no peers watching removes the fear of being judged. Learners try things they would never try in front of a group.
Make it easy to start again. If a conversation goes badly, learners should be able to restart and try a different approach straight away. Getting it wrong becomes part of learning, not a failure.
Test with real learners, then adjust. Watch how people respond to the first version. If they’re shutting down or rushing to finish, dial it back. If it feels too easy, add challenge.
At GenLearn, we build these principles in. Practice conversations are anonymous, and learners can rehearse as many times as they need, in private, before the conversation that counts. For more on testing, see how guardrails keep AI role-play safe.
Pitching the challenge for early-stage students
In a pilot with Canberra Institute of Technology (CIT), Community Work students practised with five voice-based AI client personas. CIT educators built the personas themselves, aligned to their unit requirements, with no coding or prompt design.
The students were early in their qualification, so the educators deliberately set the scenarios to be a bit easier than they would for more experienced learners. There was still plenty of challenge and realism: every persona was built on trauma-informed, strengths-based practice. They just weren’t designed to overwhelm.
The aim wasn’t to make practice easy. It was to make it achievable, so students built the confidence to take on tougher conversations later in their course, and eventually on placement.
Read more about the pilot in the Future Skills Organisation case study.
Try it for yourself
The best way to judge the balance between realism and safety is to feel it. Try a free practice conversation on our website: no one’s watching, and you can start again as often as you like.
Then think about one difficult conversation your learners face. How real does it need to feel for them to learn from it, and what would make it safe enough for them to stay?