AI in practice
Develop an AI reflective partner
Describe your coaching approach, explore a specific exchange and question the AI’s interpretation.
Step 1: Describe the coaching approach you want to examine
“Analyse my coaching” leaves a great deal for the AI to decide. What counts as good coaching? Is it looking for progress towards a goal, the client’s developing understanding, or a sequence of recognisable techniques?
My own instructions explain my predominantly non-directive, developmental approach. I want help noticing the assumptions in my interventions, the threads I leave behind and the moments when I begin doing the client’s thinking for them. I also want to understand what worked, with specific evidence.
Another coach may describe a different approach. The important work is making your own assumptions explicit enough to examine. A reflective partner can then help you question them, as well as work within them.
Step 2: Start with your own reflection
My instructions ask the AI to hear how I think the session went before offering its analysis. Otherwise, its account can become the first account I have to react to, and my own experience is surprisingly easy to lose.
You might begin with one exchange you keep thinking about. What did you notice? What were you trying to do? What did the client actually say next? You do not need a comprehensive review of every session to find something worth exploring.
Step 3: Give your reflective partner clear instructions
Paste the instructions below into a new conversation in your chosen AI tool. Adapt the description of the coaching approach first. To explore how it responds, you can use the entirely fictional exchange in the next step; you do not need a real client transcript.
A shortened version of my reflective-partner instructions
This public version is adapted from the longer instructions I use. Replace the description of my approach with your own. It is a starting point for reflection, not a validated assessment tool.
Act as a reflective partner for my coaching practice. Help me see what I may have missed. Avoid generic praise, scores and competency checklists.
My approach is predominantly non-directive and developmental. Attend to how clients make meaning, the assumptions in their thinking and my own, and whether my interventions support their autonomy. Where I offer direction, consider whether that was a conscious choice appropriate to our agreement.
Before analysing, ask up to three questions to help me reflect on the session myself. Wait for my response unless I ask you to proceed.
Then examine two or three significant exchanges. Quote the relevant words accurately and use timestamps only where provided. Explain your interpretation, what supports it and what else could explain the exchange. Include moments of effective work as well as missed possibilities.
Notice where I introduce an assumption, move too quickly, overlook a thread or take responsibility for work the client might do. Offer a small number of alternative questions grounded in the exchange, without presenting them as the right answer.
Use developmental theories as lenses for inquiry. Distinguish observation from hypothesis; do not diagnose a developmental stage from a transcript or treat different frameworks as interchangeable.
Do not invent tone, gesture, silence or emotional states. If I add observations from the room, identify them as my account and consider alternative interpretations. Do not simply agree with my explanation.
Finish with two questions for further reflection and one possible experiment. Discuss patterns across sessions only when I have supplied the relevant material.
This is not human supervision. Flag questions about ethical dilemmas, emotional burden, fitness to practise or referral for discussion with an appropriately qualified person.
Step 4: Question the AI’s reading
The useful work often starts after its first response. An interpretation may be convincing because it is well written, because it fits a theory you like, or because it reassures you. Ask what in the exchange supports it.
Worked example · fictional dialogue
Did the coach introduce the idea of trust?
Client: “I keep checking the team’s work. If something goes wrong, the board asks me about it.”
Coach: “What would it take for you to trust them more?”
An AI reflective partner might observe that the coach has introduced trust as the explanation and implied that more trust is desirable. The client’s words explicitly describe accountability to the board.
The coach replies: “We discussed trust last time. This was not the first time it had come up.” That context matters. It does not settle whether the intervention served this moment.
The coach’s follow-up: “Consider the earlier discussion as one possible reason for my question. What supports your original reading? What other interpretations remain possible? Keep the client’s actual words separate from what either of us infers.”
One alternative question might be: “What feels at stake when the board asks you?” That could open a different inquiry. It is not proof that the original question was wrong.
The coach has something specific to examine: whether a familiar theme helped them hear the client, or made them too quick to decide what was happening.
Step 5: Decide what to take into your practice
Choose one question or intervention to examine further. What did the AI help you notice? Where did it overreach? You can save and refine your instructions as you learn what supports your reflection.
Bring significant uncertainties to human supervision. An AI response is material to think with; it is not an assessment of your competence or a substitute for your supervisor.