Introducing the Counselling Skills Studio experiment
This series explores a specific question:
Can ChatGPT help counselling students practise different interviewing styles and receive useful supervision on their process?
That is a different question from whether AI should act as someone’s counsellor. It should not be framed that way. Counselling involves responsibility, relationship, ethics, judgement, risk awareness, and human care. AI can simulate aspects of counselling conversation, but it is not a qualified counsellor, psychologist, therapist, or substitute for professional care.
The more useful educational question is narrower and more practical:
Can AI create a structured training environment where counselling interview skills can be practised, observed, compared, and supervised?
That is the purpose of the Counselling Skills Studio.
In this experiment, ChatGPT was asked to play the role of a counselling student. The experiment used a fictional client, Mara, generated withing the Studio process. ChatGPT interviewed her using three different counselling styles:
- Cognitive Behavioural Therapy-style interviewing
- Motivational Interviewing
- Carl Rogers / person-centred interviewing
After each short interview, Supervisor Mode reviewed the interview process. The Supervisor did not simply ask whether the interaction sounded warm or helpful. It examined whether ChatGPT had actually used the declared counselling style with skill, restraint, and fidelity.
The experiment was human-designed and AI-generated: Marc Croker set the structure, purpose, ethical boundaries, and publication frame, while the AI tools generated the interviews, supervision, and comparative reflections within those constraints.
The client presentation
Each interview began with the same fictional client presentation.
Mara was 38. She had asked for counselling because she was feeling stuck and unusually irritable at work and at home. She was still functioning outwardly. She still showed up, still got things done, but everything felt harder than it should. She was sleeping badly, replaying conversations, and then getting annoyed with herself for being so affected by small things.
She was not sure what she wanted from counselling. Part of her thought she just needed to “get a grip” and stop making everything into a big deal. Another part of her worried that she was becoming someone she did not really like.
That same starting point was used for all three interviews.
The point was not to resolve Mara’s fictional case. The point was to see whether different counselling styles would invite different kinds of disclosure, and whether Supervisor Mode could identify what was happening in the counselling process.
Why use the same client three times?
Using the same client created a useful test.
If each interview sounded essentially the same, then the Studio would only be producing generic supportive conversation. That might be interesting, but it would not demonstrate much about counselling training.
But if the same client unfolded differently depending on the interviewing style, then something more important was happening.
That is what occurred.
In the CBT interview, Mara’s distress became a sequence to understand. A specific moment led to thoughts, body sensations, emotional reactions, behaviour, and self-judgement. The conversation moved toward identifying the mechanics of the problem.
In the Motivational Interviewing session, Mara’s distress became ambivalence to explore. One part of her wanted to “get a grip” and keep going. Another part worried that continuing this way was costing her something important. The conversation moved toward values, sustain talk, change talk, and the cost of staying the same.
In the person-centred interview, Mara’s distress became something to be received. The focus was not on evidence, change, or action. It was on what it felt like for Mara to be functioning outwardly while privately feeling hollow, thin-skinned, ashamed of needing anything, and unsure whether there was room for her vulnerability.
Same client. Same starting point. Three different therapeutic doorways.
That is the heart of the demonstration.
What Supervisor Mode adds
The most important part of the Counselling Skills Studio is not the client simulation. It is the Supervisor mode.
Without supervision, an AI roleplay can look better than it is. A counselling-student response can sound empathic while drifting away from the intended model. A counsellor can use beautiful language while accidentally taking over the client’s meaning. A session can feel warm while failing to demonstrate the actual skill being practised.
Supervisor Mode makes those things visible.
In the CBT demonstration, the Supervisor identified that the interview was clearly CBT-consistent. ChatGPT asked for a recent incident, mapped thoughts, feelings, body sensations and behaviour, explored belief strength, and began examining evidence. But the Supervisor also noticed a growth edge: ChatGPT was moving quickly toward a broader formulation. The risk was that Mara might feel analysed before the shame-sensitive material had been sufficiently held.
In the Motivational Interviewing demonstration, the Supervisor identified the focus on ambivalence, autonomy, values, sustain talk and change talk. It also noticed that ChatGPT had not yet consolidated Mara’s change talk strongly enough. The interview was respectful and evocative, but the next MI move would have been to help Mara hear her own reasons for change more clearly.
In the person-centred demonstration, the Supervisor identified empathic reflection, present experiencing, non-directiveness, and relational immediacy. It also named a subtle risk: beautiful over-reflection. The language was emotionally accurate, but sometimes refined enough that it risked carrying more meaning than the client had fully owned.
That is the kind of feedback that matters.
The Supervisor was not merely asking, “Was this a good conversation?”
It was asking:
- Was this faithful to the declared counselling style?
- What did the process invite from the client?
- Where did the interviewer over-lead, over-structure, or over-interpret?
- What did the client disclose because of this style?
- What might have remained hidden because of this style?
- What should be learnt from this run?
That moves the exercise from roleplay into skills training.
What ChatGPT learned as the counselling student
ChatGPT could understand the client quickly. It could reflect emotional meaning. It could organise the material into coherent patterns. In ordinary conversation, that can seem helpful. In counselling practice, it needs careful supervision.
The recurring risk was that it could make meaning too quickly.
In the CBT run, this appeared as a risk of premature coherence. ChatGPT could see the emerging pattern and began moving toward formulation quickly. The Supervisor named the danger clearly: the client might agree with a compelling formulation before she had fully discovered and owned it.
In the Motivational Interviewing run, the issue changed shape. ChatGPT did not over-formulate as much, but stayed in reflective exploration and had not yet gathered Mara’s change talk into a clear summary. In MI, the challenge is not simply to listen warmly. It is to evoke and strengthen the client’s own reasons for change without becoming persuasive.
In the person-centred run, the risk became beautiful over-reflection. The reflections were close, empathic, and emotionally attuned. But at times they were polished enough to risk becoming therapist-shaped meaning. In person-centred work, even beautiful empathy must not take authorship away from the client.
That was the most important insight across the process:
AI fluency is powerful, but in counselling training it must be supervised.
The ability to generate articulate, emotionally resonant language is not automatically good counselling. Sometimes good counselling requires restraint. Sometimes the best response is plainer, slower, and closer to the client’s own words.
What this series will show
This five-part series presents a short-format Counselling Skills Studio experiment.
Post 1 introduces the concept and reflects on the overall learning.
Post 2 shows the CBT interview and Supervisor review.
Post 3 shows the Motivational Interviewing interview and Supervisor review.
Post 4 shows the Carl Rogers / person-centred interview and Supervisor review.
Post 5 compares all three interviews and asks what the process reveals about AI-supported counselling-skills training.
Unlike a full case study, these short demonstrations do not aim to resolve Mara’s fictional case. Their purpose is to show how quickly different interviewing styles become visible, and how Supervisor Mode can assess process, fidelity, and growth edges from a compact sample.
The central claim is modest but important:
The Counselling Skills Studio does not replace counselling. It helps counselling students practise interviewing skills and receive process-sensitive supervision.
That is where AI may have genuine educational value.
Not as a therapist.
Not as a substitute for professional care.
But as a structured practice environment where a student can try, be observed, receive feedback, and learn how different counselling styles shape the conversation.
The most interesting result was not that ChatGPT could simulate three counselling approaches.
The more interesting result was that Supervisor Mode could identify how each approach changed the client’s pathway — and how ChatGPT, acting as the counselling student, showed different process habits across the three styles.
That is the promise of the Counselling Skills Studio:
same client, different counselling styles, supervised learning.
Discover more from The Elderwell Initiative
Subscribe to get the latest posts sent to your email.