How to use ChatGPT (and Claude) in your coaching practice

Helping thoughtful, experienced coaches use AI in ways that deepen their practice and preserve their judgement.

You may have opened ChatGPT, asked a few questions and wondered how it could contribute to your coaching. A useful next step is to work with it on something you recognise: preparing for a session, developing a client summary or examining an intervention you are unsure about.

This article introduces those uses and the decisions behind them.

My starting point is augmentation: using AI to build on your coaching experience while continuing to exercise your own judgement. Its usefulness depends on the purpose you choose, the information you provide and how you examine what comes back.

About me: I have been an executive coach since 2015, working with founders, CEOs and senior leaders. My experiments with AI in my own practice became the AI-Augmented Coaching course.

What can ChatGPT actually do for a coach?

ChatGPT and Claude are AI assistants you work with through a conversation. You describe a task, give the assistant appropriate information, examine its response and ask for changes. You can begin with an ordinary chat; you do not need to build a separate application.

Three uses are worth exploring:

• Session summaries. Use the written transcript of a call to develop a draft you check before sharing.
• Reflective practice. Examine an exchange, an intervention or a question about your own coaching.
• Preparation. Revisit previous notes and distinguish the client’s open questions from actions they actually agreed.

Choose the task that addresses a need in your practice. Your own notes may be enough for preparation; a fictional exchange is enough to begin learning about reflection. Recording every client conversation is not a prerequisite for getting started.

How do you use ChatGPT to transcribe and summarise coaching sessions?

Before recording a client or uploading their words, work through the data and confidentiality decisions. Client agreement, suitable provider arrangements and a clear purpose all need attention. You can learn the process with an agreed practice call or entirely fictional material.

First distinguish the records. A recording preserves audio or video. A transcript turns speech into written text. A summary selects and interprets parts of that text. A meeting service such as Zoom, or a note-taking tool such as Fathom or Granola, can provide the transcript, depending on the setup you choose.

You then give the transcript to an AI assistant with instructions about who the summary is for and what it should include. For a client summary, you might ask for their important insights, questions left open and next steps they explicitly chose. Ask it to preserve uncertainty and flag wording that needs checking.

Read the draft against the transcript, your notes and your recollection. A transcript can mishear words; a summary can turn a possibility into a commitment. Use a recording to investigate disputed wording where one is available. Add observations from the session carefully, distinguishing what you noticed from what you inferred.

Keep discussing the draft until you have an account you can stand behind, then review the email before sending it.

Also ask whether a summary serves this client. Some value a record; others prefer to make their own notes or leave a question open.

Can ChatGPT be a reflective partner for your coaching?

An AI reflective partner is an assistant you have instructed to help you examine your coaching. You explain your approach and the question you want to explore, then discuss its reading of an exchange or your notes.

My approach is predominantly non-directive and developmental. I want to examine what my questions introduce, where I begin doing the client’s thinking and what I may have left unexplored. Another coach may want to attend to different things.

Write down your own reflection first. Then ask the assistant to distinguish observations from interpretations, refer to the relevant words and consider alternatives. It should hear your account before producing an assessment for you to react to.

Instructions can encourage useful challenge, but they cannot guarantee it. Generic praise offers little to examine; a harsh verdict is no better simply because it sounds demanding. If the assistant says you imposed an agenda, ask what supports that interpretation, what context is missing and what else could explain the exchange.

Notice how your own description shapes the answer. An assistant given your favourite framework may keep interpreting everything through it. Ask it to reconsider without that framework and compare what changes.

What about ChatGPT prompts for coaching?

A prompt is simply the instructions you give the assistant. A useful example can help you begin. The learning comes from understanding what those instructions ask for, adapting them to your purpose and examining the result.

Describe the job, the relevant material, who the output is for and how you want uncertainty handled. For reflection, explain your coaching approach and what you want help noticing. You can use ordinary language.

Continue the conversation. Ask it to shorten an answer, return to a passage or consider another explanation. If it is too approving, ask for evidence and alternatives. If it produces a list of faults, bring it back to one intervention and what can reasonably be inferred.

Keep a dated copy of instructions that work for you, separate from client material.

Is it safe to put client information into ChatGPT or Claude?

Paying for a subscription does not, by itself, make the account suitable for client material. A paid personal account and an account supplied under business terms can have different contracts, even when their chat windows look similar.

Check the exact service, plan and features you intend to use. Where the provider processes personal data on your behalf, the necessary processor contract must cover that use. Training settings are a separate question, alongside access, retention, deletion, overseas processing and any other providers involved.

Client agreement matters too. Explain the particular activities and record the choices: recording, transcription, a client summary and reflection on your coaching are distinct uses. A general signature or permission to use AI does not resolve every data-protection obligation.

Removing names is not necessarily anonymisation. A distinctive role, event or combination of details may identify the person. Your own reflective notes can also contain personal information, even if you never attach a transcript.

Begin with fictional material while you work through the arrangements for real client information.

Claude vs ChatGPT for coaching, which should you use?

I use Claude extensively in my own practice, and work with both Claude and ChatGPT on my course. Your choice should depend on the task, the quality of the work you can produce together and the terms of the account you would use.

Try the same short fictional exchange with the same instructions in each. Form your own view first, then compare the responses. Which preserves what was actually said? Which separates evidence from interpretation? What happens when you challenge a conclusion or supply missing context?

Consider the practical fit as well: how easy it is to work with files, reuse instructions, find earlier work and manage the settings you need. Check current features, usage limits, prices and account terms. A product name alone does not answer those questions.

Record which model and account you tried, and when. One useful answer is a starting point for learning, not proof that a tool will handle every task well. Two assistants agreeing is not independent verification of an interpretation.

You can start with one tool and become familiar with how to work with it.

Where AI shouldn’t reach

Decide what you remain responsible for: understanding the agreement with your client, examining what the tool produces and choosing what to share or act on. An assistant can contribute to that work, but a convincing response does not settle those decisions.

I distinguish cognitive offloading from cognitive surrender. Having AI gather passages or organise notes may free attention for reflection. Accepting its account of the client or its verdict on your coaching without examining it gives away a different part of the work.

There is also a question of volume. If a short reflection turns into pages of analysis, choose what deserves attention and what you can leave. More output is not automatically better coaching.

Giving a client an AI tool to use between sessions introduces a different arrangement, with its own purpose and responsibilities.

How do you learn this properly?

Start with one familiar task and fictional practice material, and put the data and contracting arrangements in place before moving into real client work.

On my live AI-Augmented Coaching course, experienced coaches develop these ways of working together. Demonstrations, supporting materials, experiments between sessions and discussion give you opportunities to compare experiences and question assumptions. The aim is an approach you understand, can explain and can revise as you learn.