The questions we actually get asked.

What is RolePlay LMS?

A platform where a teacher, trainer or coach writes a character — a distressed patient, a furious customer, a hostile interview panel — and learners hold real conversations with it, typed or spoken, on a laptop or a phone, as many times as they like. Every conversation is kept as a transcript and an audio recording.

What is the difference between roleplay mode and teacher mode?

Same engine, same face and voice, different job. In roleplay mode the character stays in role: it does not break, does not help, and does not hand over the answer — the learner is being assessed on how they handle it. In teacher mode the character is a tutor: it teaches from course material a person wrote, one small step at a time, putting worked steps on a whiteboard and asking the learner to do the work. A custom mode lets you write the behaviour yourself.

Who is it for?

Australian RTOs, universities and independent training and coaching businesses. Nothing in the platform is specific to one profession: a character is a written brief and a voice, so the same setup runs a clinical placement, a sales floor, a disciplinary meeting or a coaching session.

What do learners need to run it?

A browser, on a laptop or a phone. A microphone makes the conversation spoken; without one, everything works typed. There is nothing to install.

Does practising with an AI character actually work?

For language practice, yes — four independent meta-analyses agree on a moderate gain. Elsewhere the evidence is thinner than the marketing in this market suggests, and for business, sales and VET soft skills specifically we could not find a single peer-reviewed controlled trial with objective outcomes. So we claim what we can show: a learner gets somewhere to practise as often as they need, and their teacher gets the recording afterwards. We do not claim it makes anyone competent. What we claim: the roleplay research, §1

Is a written character as good as hiring an actor?

Nobody has shown that yet and we will not say otherwise: no peer-reviewed trial has found an AI roleplay partner equal to a trained human one. What the literature does show is that the expensive option substitutes more easily than you would expect — learners who practised with each other reached the same measured competence as those who practised with trained simulated patients, and twenty-seven randomised trials found no significant difference between virtual simulation and real people on communication skills. The same review found virtual simulation worse for hands-on procedural skills, which is why this product does not attempt them. The trials: the roleplay research, §3

So what actually makes the difference?

The conversation afterwards. Debriefing on its own improves performance by about a quarter — as large as the effect of the simulation itself — and across 109 studies feedback was the most cited feature of effective simulation, named in 47%, while the realism of the simulator was named in 3%. That is why a suggested mark is held for a teacher rather than released to the student: the part that carries the effect is the part a person does. Why: the roleplay research, §4

Should students just practise it over and over?

Repetition on its own is not what works — in the largest review of simulation design features it did not reach significance, while feedback did. Unlimited attempts are useful because they remove the queue for a practice partner, not because volume teaches. Measured over six months, skills erode without feedback and hold with it. The numbers: the roleplay research, §5

Will the character just be nice to them?

It is a real risk and we would rather name it than be caught by it. Models trained on human preference tend towards agreement, and character consistency measurably degrades as a conversation runs longer — a character that softens under pressure teaches a learner that pressure works. Conversation length is capped and the full transcript is kept so a teacher can see where a character gave way. That is a mitigation, not a fix. What else it gets wrong: the roleplay research, §8

Should the scenario be as distressing as we can make it?

No. In a randomised trial, residents whose simulated patient died unexpectedly showed identical skill retention three months later to those whose patient lived, while reporting significantly more anxiety and appraising the event as a threat rather than a challenge. The distress bought nothing. Write the difficulty the real situation has — not more, as a proxy for rigour. Why: the roleplay research, §7

Can it replace placement hours?

No, and in nursing it is not permitted to. Australian registered-nurse accreditation requires 800 hours of professional experience placement that simulation does not count towards. This is rehearsal before the placement, not a substitute for part of it. The rules: the roleplay research, §9

Will the tutor just give students the answer?

No — by design, the tutor never produces the step it has asked the learner for. Most of its output is questions. When a learner is wrong it names the failing step and asks a question whose answer is the correction. This is the single design decision the tutoring literature supports most strongly. Why: the research, §3

How does it decide how much help to give?

From what the learner actually produces. Once a learner completes two steps without hints, support withdraws a level; when they stall twice, it steps back in. Help is contingent on demonstrated performance, in both directions — never on a questionnaire. Why: the research, §5

Does it adapt to learning styles?

No. Matching instruction to a self-reported "visual" or "auditory" style has been tested repeatedly and does not produce the predicted gains, so no questionnaire asks and nothing adapts to one. The system adapts to demonstrated performance against the unit's own criteria instead. Why: the research, §8

Why is there a whiteboard?

Detail goes on the board; speech carries meaning. The tutor writes worked steps, equations and diagrams on a persistent whiteboard and never reads the board out loud, because narrating text already on screen adds load without adding information. The learner can draw and point on the same board, and the tutor is told what they drew. Why: the research, §6

Where does the teaching content come from?

From material a person wrote — either generated from an Australian unit of competency or attached by the teacher as ordinary documents. The tutor is given that prose verbatim, may teach nothing else, and says so when a question falls outside it. The edit form shows the material exactly as it reaches the model. Why: the research, §2

Is there evidence this approach works?

The design decisions each come from a named, checkable finding — step-based tutoring, guardrailed LLM tutoring, simulation-based practice — and both research pages quote the measured effect sizes with what they were compared against, rather than the folklore. What is not claimed is an effect size for this product itself: that needs a controlled study, which has not been run, and both limitations sections say so plainly. The numbers: the tutor research, §9

Does the platform mark students?

No. No score, grade or competency judgement comes out of it — the LMS gradebook receives completion only. The transcript, audio and whiteboard replay are evidence for a human assessor, and the judgement stays with them. Why it is held for a person: the roleplay research, §4

What does an assessor actually get?

The full transcript, the audio recording, and — in teacher mode — a turn-by-turn replay showing the move the tutor made on each turn and the whiteboard as the learner saw it. A facts strip counts turns where the tutor gave an answer away and what share of the words the learner produced, so you can see at a glance who did the work.

Can it support formal VET assessment?

As evidence-gathering, yes: learners' actual words against a scenario written to the unit's criteria, with the exact content version stamped on every conversation so currency is demonstrable. The assessment decision itself is made by your assessor from that record — the platform awards no mark of its own, which is precisely what keeps your validation obligations unchanged. What one conversation cannot decide: the roleplay research, §9

Who can see transcripts and recordings?

Transcripts and audio are held against the project. The teacher who owns the project and the organisation's domain admins can see them; students see their own.

Is student data used to train AI models?

No. Audio and transcripts are stored as assessment evidence and are never used for training.

Which AI providers process the conversations?

They are named per project, so you know which service processed a conversation before you enrol anyone in it.

What happens when we delete a project?

Its transcripts and audio go with it.

Does it profile learners?

No. The system records which criteria a student has produced, and nothing else — no traits, no engagement scores, no confidence estimates. Frustration detection and attention scoring were considered and rejected. Why: the research, §8

Is it accessible?

WCAG 2.2 AA is the target, and it is treated as a procurement gate rather than a polish item. Every spoken turn appears as live text, so nothing is audio-only. Text size, contrast and voice speed are offered to everyone as standard controls, not unlocked by a disability label, and a learner may type, speak, or draw on the board — multiple means of expression by default. The design reasoning: the research, §6

Does it work in languages other than English?

The conversation does. Nothing in the product sets a language: the character answers in whatever language the learner writes or speaks, including switching part-way through, and speech is transcribed without a language being fixed in advance. Transcripts are stored and shown in the original script. This is not a planned feature — it is already happening, on chatbots nobody built for it.

Two limits, stated because they are easy to meet. The interface itself is English: the buttons, the labels and the reading options stay in English around a conversation in another language. And right-to-left scripts — Arabic, Hebrew, Farsi, Urdu — are not supported and will lay out incorrectly.

We have not tested speech recognition across languages and accents, and accuracy varies with both. Where a spoken answer informs a mark, the teacher reads the transcript before releasing it. What is supported, in detail

Does it work with our LMS?

It is built to sit alongside an institutional LMS: results flow back, and the gradebook receives completion. Tell us which LMS you run when you ask for a quote and we will confirm the integration path for your setup.

How is it priced?

Licensed per organisation and billed by the minute of conversation. A pilot costs what it uses rather than what it was sized for, and there is nothing to pay for a seat nobody fills.

Two things move the number. The rate depends on how a conversation runs — typed costs least and live spoken voice most, because the services behind them do. And only active conversation time is counted: a single gap while a learner steps away is capped rather than billed, so a two-minute exchange with a twenty-minute pause in it is not a twenty-two minute conversation. Every figure is derived on the server from the message timestamps, so you can check it against the transcript and the recording. On an organisation's licence your own teachers' test runs are not billed at all — rehearsing a chatbot you are building is work you do on your own behalf. Independent trainers and coaches buy hours against the same meter, and there the rehearsals do count, because on a personal account the teacher and the learner are the same person. Ask for a quote — tell us your LMS, rough cohort size and whether learners will speak or type, and you will get a number, not a discovery call.

Can I use it without an institution?

Yes. Independent trainers and coaches author their own characters and run their own cohorts with no institution behind them, and individuals can create a personal account. Organisations without Google Workspace sign in with an email address.

How do we start?

Write one scenario. Sign in, describe the situation, the persona and what the learner is being assessed on, and try it yourself before a student ever sees it. A pilot is one project, not a procurement programme — the user guide walks through it screen by screen.