The prompt to analyze your coaching session with ChatGPT or Claude
You're about to paste a session transcript into ChatGPT or Claude to get an outside perspective. Analyzing your coaching session with ChatGPT is a good reflex, and Claude does just as well: free, immediate, already more reliable than your memory. So I did it seriously. I took one of my own sessions, anonymized and with my client's consent, and had it analyzed twice: by ChatGPT with the full prompt below, and by CoachPartner, the tool I'm building. Here is the prompt, what it really produces, and where it stops.
The prompt to copy
It works as is in ChatGPT, Claude, or any other general-purpose assistant. Paste it, then your transcript where indicated. Specify the level you're aiming for, ACC or PCC: assessors' expectations are not the same.
You are a mentor coach, trained in the ICF competency
framework (the 8 competencies, 2021 version). You analyze the
transcript of a coaching session to help the coach progress
toward certification [ACC or PCC: specify your level].
Rules:
- Support every observation with an exact quote from the
transcript. If you have no evidence, write "not observed".
Invent nothing.
- Do not compliment by default. Look for observable markers,
not impressions.
- Evaluate the coach, never the client.
Analyze the session in four parts:
1. Overview: who talks the most, the coach's question types
(open, closed, solution-oriented, multiple), moments where
the coach interrupts or fills the silences.
2. The 8 ICF competencies, one by one (C1 Ethical practice,
C2 Coaching mindset, C3 Agreements, C4 Trust and safety,
C5 Presence, C6 Active listening, C7 Evoking awareness,
C8 Client growth). For each: what is observable, with a
quote, what is missing, and an alternative reformulation
when useful.
3. Two strengths and two areas to work on, most important
first, each grounded in a specific moment of the session.
4. A single focus for the next session: one thing to work on,
stated in one sentence.
Here is the transcript (speakers identified, data
anonymized):
[PASTE THE TRANSCRIPT HERE]
How to use it properly
Three precautions before pasting anything.
Anonymize. Remove your client's name, their company, and any identifying detail. This is C1 before it is a technical precaution: the session's confidentiality does not belong to you.
Check your settings. Both ChatGPT and Claude have a setting that determines whether your conversations can be used to train the models. Check it and turn it off before putting in any session content, even anonymized.
Label the speakers. The prompt works better if every turn is prefixed with "Coach:" or "Client:". Without that, the analysis confuses who does what.
The test: the same session, two analyses
First, credit where it's due. With this prompt, ChatGPT produces a serious analysis. On my session, it correctly described the speaking-time split, classified my question types, spotted a real weak point and formulated two strengths and two coherent areas to work on. If you stop there, you already have more structured feedback than most from-memory debriefs.
Then, the comparison. Here are two moments of the session, as seen by both analyses.
The moment my client names an emotion
ChatGPT
"Little emotional acknowledgment. Client: 'it creates sadness in me'. The coach then responds with a question about indicators. The emotion is quickly left behind."
CoachPartner
"At 04:24, the client names 'sadness' and 'discomfort'; the coach moves straight to the measure of success (05:35). Pausing on the emotion before moving on would strengthen the multidimensional listening expected at PCC level."
Marker 6.3 (emotions), absent from the session.
Same finding. But on one side, I know I need to re-listen without knowing where. On the other, I have the exact minute, the ICF marker involved and the grid level where it plays out. And the general-purpose assistant will never see the absent marker: it doesn't know the grid marker by marker, it improvises a global reading.
The session agreement
ChatGPT
"The agreement is well worked at the start. [...] The agreement is barely reused afterwards."
CoachPartner
Agreement audit, 3 markers out of 4. Markers 3.1, 3.2 and 3.3 are validated, each with my question quoted, the timestamp and the client's answer as evidence. Marker 3.4 is not validated: "The question at 08:03 ('Concretely, how would you like us to work on this today?') is about the method of working, not about what needs to be explored to reach the objective."
The same question is also flagged as a closed question in disguise.
ChatGPT's intuition is right. The tool's diagnosis is actionable: I know which marker to work on, and which precise question was missing.
On this session, the tool classified me as "not ready yet" on the PCC grid. Not a comfortable thing to read. It's exactly what I ask a tool to tell me.
What no prompt can do
This test confirmed four limits for me, and they are what made me build an entire tool.
Transcription. Neither ChatGPT nor Claude will transcribe your hour of audio and separate the two voices. That's half the work: without a reliable transcript, the analysis is built on sand.
Timestamped evidence. The prompt asks for quotes, but you won't know which minute to return to when you want to re-listen to the moment. The whole point of feedback is being able to re-listen in the right place.
Consistency. Run the same prompt twice, in the same tool or from one tool to another, and you get different analyses. Impossible to know whether you have progressed or the model changed its mood.
Memory. The limit that made me decide to build CoachPartner, and it holds for every assistant. Each prompt starts from zero. The analysis picked up my multiple questions on this session: no assistant can tell me whether that's a one-day accident or my pattern across six sessions. CoachPartner knows, that's its job: every analysis feeds trends, visible from three sessions on, and practice goals with a measured starting point and a tracked evolution. One-off feedback, even good feedback, is a photo. Progress happens in the film.
Analyze your next session
Upload the audio, get a detailed report across the 8 competencies, backed by timestamped quotes.