What is an AI practice client? Simulation versus being coached by AI

An AI practice client is software playing a fictional client so that a human can practise coaching. You are the coach; the software is pretending to be the person in the other chair.

Being coached by AI reverses those roles; it may use a different kind of software. There, the software asks the questions and you are the one working on something.

Both are commonly described as "AI coaching", which is why people end up comparing products that do opposite things.

Who is coaching whom

Automated feedback comments on coaching responses. CARE, for example, offered feedback during simulated practice (CHI 2026). A system’s comments are neither the client’s lived experience nor professional supervision or certification.

There's a fourth arrangement that involves no AI at all: two coaches taking turns, one bringing something real. Qualifying peer exchanges can count under ICF's client-experience rules; other bodies set their own conditions.

What each looks like in the research

In one study, counsellor trainees pasted a role-play prompt asking ChatGPT to act as a fictional client named Omar, then coached in text for 30 to 40 minutes (Behavioral Sciences, 2025). The human was practising; the model held the client role. That is a practice client.

In another, a chatbot called Vici asked participants about their goals, tested how realistic those goals were and helped them build an action plan over ten months (PLOS ONE, 2022). The human was the client. That is AI coaching.

The studies address different roles. Evidence of AI-delivered coaching does not establish that practising with an AI client improves a human coach's work.

What a persona actually is

A fictional persona combines instructions with what the model generates in response to you. Its described feelings and history belong to the roleplay; they are not reports from a person having those experiences.

The trainees in the 2025 study were precise about how that shows up. The client agreed too readily: "It was very agreeable... it just went with what I was saying." The transcript analysis found complete sentences, no slang and few fillers, which read as polished rather than spontaneous. It arrived culturally blank unless someone specified otherwise: "Naturally, it's very neutral. If you want culture or stereotypes, you have to be very specific." And in some sessions it resolved everything: "In just a few prompts, I solved all their problems."

One trainee observed this while reviewing a moment where her AI client appeared to reach an insight. "I may have viewed it... as validating me. I felt like it was more a part of its agreeableness, as opposed to a client's self-awareness." A simulated client can produce the sentence a real client might say after a breakthrough without anything having happened.

A session that felt wonderful does not establish coaching competence; accommodating model behaviour may explain it.

Where the products sit

Commercial tools cover both directions, so check the roles before comparing them. Look separately for a fictional client, feedback on your coaching, and support for your own goals. A tool may offer more than one function.

Robin, which publishes this guide, describes spoken rehearsal with a practice client called Sarah, followed by a debrief. Its AI Practice page says, “Robin does not grade your coaching or certify that you are ready.” As checked on 11 September 2026, live rehearsal requires Pro + AI for new subscriptions; existing unlimited subscribers retain their agreed access. Check that page for current access and usage limits. This is a provider’s description of its product, not evidence that using it improves coaching.

When you're weighing up any of these, five questions sort out what you're being sold:

1. Who has the objective in the session, you or the software?
2. Who is doing the coaching?
3. Is the topic real or fictional?
4. Where does any feedback come from, and what is it comparing you against?
5. What exactly is being claimed, and what was measured to support it?

Confidence scores, mastery tracking and framework references describe what a product does; they are not findings about whether your coaching changed.

Two boundaries to keep

Hours. The ICF defines a client coaching hour as 60 minutes with a client who has hired you specifically as their coach, in real time; shorter sessions count proportionately (ICF experience requirements). It addresses paid work, pro bono work and barter. Its experience page does not establish eligibility for coaching an AI simulation. Keep rehearsal separate from client-experience hours and check the current rules of the body you’re applying to.

Privacy. Invent the client rather than changing a real client's name or using a friend's situation. Whatever a provider's data controls say, a real person's material does not belong in a rehearsal.

Before choosing a tool, ask its provider what evidence supports the function you want: fictional-client rehearsal, feedback on your coaching, or help with your own goal.

Sources

• CARE: Can LLM-Simulated Practice and Feedback Upskill Human Counselors?, CHI 2026, DOI 10.1145/3772318.3791821. ArXiv v2, 9 March 2026: optional feedback during simulated counselling practice; not credential assessment.
• Learning Through Simulation: Counselor Trainees' Interactions with ChatGPT as a Client, Behavioral Sciences, 2025. Nine submitted transcripts, four focus-group participants, one fictional case; counsellor trainees.
• Terblanche, Molyn, de Haan and Nilsson, Comparing artificial intelligence and human coaching goal attainment efficacy, PLOS ONE 17(6): e0270255, 2022. AI delivering coaching to human participants.
• ICF coaching experience requirements, checked 11 September 2026. ICF credentials only.
• Robin, hear a practice call, checked 11 September 2026. Publisher's own product description.