ChatGPT Prompts for Executive Coaching: Prep, Reflection and Goals

Executive coaches get specific session prep from ChatGPT by giving it context first: the client's role and situation described without identifying details, the goal, what has been tried, stakeholder feedback themes and what the client wants from the session. Then ask for questions, hypotheses to test and structures to offer, not advice to pass on.

The generic output comes from the generic input. "Give me coaching questions about delegation" returns the questions every coaching website has published. The fix is a context block you write once per client, prompts that draw on it, and a confidentiality line you hold while doing so, with the judgement left where it belongs: with you, in the room.

Follow me on Instagram@sagnikteaches

The context block: what to supply before any prompt

Create one reusable block for each client and paste it at the start of a session-prep chat. It takes about five minutes to write and lifts every prompt that follows. Keep it anonymised: roles and sectors, not names and companies.

Connect on LinkedInSagnik Bhattacharya

CONTEXT (anonymised)
Role: Operations director, about 18 months in role, promoted from within. Reports to the CEO. 40 people across three teams.
Situation: Two of her direct reports are stronger technically than she is; she tends to step in and fix problems herself. Board presentation in eight weeks.
Coaching goal agreed at contracting: become a leader who sets direction and lets the teams deliver, without losing standards.
History: Sessions 1 to 4 covered role clarity and her calendar. Last session ended with her agreeing to delegate one weekly report and to note how it felt.
Stakeholder themes (from interviews, de-identified): trusted, fast, detail-oriented; "sometimes hard to know what she wants"; "re-does work quietly".
What the client asked for today: help preparing a difficult conversation with one direct report.
My style: questioning rather than advising; I use contracting and a simple GROW structure.

Notice what is in it and what is not. Role, situation, goal, history, themes, request and your own style are there. Names, employer, sector detail and anything about the client's personal life are not. Your style line matters because it tells the model to produce questions in your manner, not a consultant's list of recommendations.

Subscribe on YouTube@sagnikteaches

Why the context lines each earn their place

• Role and tenure: a new director and a veteran need different questions about authority.
• Situation with pressure points: the board presentation in eight weeks tells the model there is a deadline shaping the session.
• Agreed goal: keeps prep tied to the contract, so a good conversation does not drift away from what the sponsor and client agreed.
• History: stops it proposing a topic you covered in session one.
• Stakeholder themes: the strongest source of useful hypotheses; without them the model reaches for stereotypes.
• Your style: turns advice into questions.

Prompt 1: session prep that is specific

Using the context above, help me prepare for today's session. Do not give the client advice. Give me:
1. Three hypotheses about what might sit underneath her request, each with the evidence for it from the context and what would make me drop it.
2. Six open questions in my style, ordered from rapport to challenge.
3. One structure I could offer for the difficult conversation she wants to prepare (for example, a way to open, the feedback, and the ask).
4. Two risks: where I might over-lead or collude with her story.

A realistic output (illustrative) might offer hypotheses such as: "She may associate stepping in with being valued, so delegating feels like becoming unnecessary" (evidence: promoted for technical strength; stakeholders say she fixes things quietly) and "She may be avoiding direct feedback with this report because the standards gap is real and she does not want to hurt the relationship". Questions might include "What would you want them to say about working with you a year from now?" and "What is the cost, to you and to them, of the version of this conversation you have been avoiding?". The structure could be: opener that names the shared goal, specific observation, impact, the report's perspective, agreed next step.

What you fix: delete hypotheses you cannot link to evidence in the context. Rewrite any question that sounds like a leading statement ("Don't you think you should...") into something genuinely open. The model's risks section is usually the least generic part, and worth reading before you walk in.

Synthesising stakeholder feedback

This is where ChatGPT saves the most time, and where the confidentiality care is greatest. Stakeholder interviews and 360 comments are given to you on the understanding that they will be handled carefully. De-identify before you paste: remove names, titles that identify one person, product or team names, and anything that makes a comment attributable ("the only person in finance who..."). If the feedback instrument or the participants' agreement does not allow use of AI tools, do not.

Prompt 2: theme synthesis you can check

Below are de-identified feedback comments about one leader, from 9 respondents. Group them into no more than five themes. For each theme: a short label, a two-line description, the number of respondents who raised it, and two verbatim quotes. Put comments that contradict each other in a separate section called "Tensions". Do not interpret motives. Do not add themes that are not in the comments.
[paste]

Sample result (illustrative): "Theme 1: reliability (7 of 9): 'always delivers', 'you can count on her in a crisis'. Theme 2: unclear direction (5 of 9): 'hard to know what she wants', 'priorities change by Thursday'. Tensions: 'very hands-on, which helps' versus 'hands-on to the point of redoing our work'."

What you check: count one theme yourself against the raw comments, read the "Tensions" because they are often the richest material, and make sure no quote could identify the commenter. The model sometimes merges two different comments into one neat quote; if a quote does not appear in your source, remove it. The synthesis is a map. The reading of what it means for this client is yours.

Prompt 3: from themes to questions, not conclusions

For each theme above, give me two coaching questions that would help the client notice the pattern in their own words, without telling them what the pattern means. Add one question for the Tensions section.

This keeps the model in the supporting role. You are not asking what the client should do; you are asking what you might ask, which suits a method built on the client reaching their own conclusions.

Turning "be more strategic" into something observable

Many coaching goals arrive as vague wishes: "be more strategic", "improve executive presence". A good prompt helps you sharpen them with the client rather than for them.

The client says her goal is "to be more strategic". Using the context, suggest five different ways this goal could be made observable, each with a measure the client and her sponsor could both see within three months. Then give me three questions that would let her choose between them herself.

Illustrative output: "Observable versions: (1) share of her week spent on planning rather than reviewing, tracked by calendar sampling; (2) number of decisions the team takes without escalating to her; (3) the board presentation delivered with direction and trade-offs rather than status; (4) one standing weekly priority-setting conversation with each report; (5) sponsor's view at the three-month review." Questions: "Which of these would you notice first? Which would your CEO notice first? Which would worry you?"

What you fix: discard measures the client cannot influence or that invite gaming, and check each against the contract with the sponsor. Where the sponsor is a third party, goals and measures should come from the contracting conversation, not from a prompt.

A filled-in goal frame

Reflection questions for between sessions

Clients often do their best thinking between sessions, and a well-built prompt helps you send questions that are tied to the session rather than generic journalling prompts.

Based on the session summary below (anonymised), write three reflection questions for the client to answer in writing before our next session. Each should be answerable in 10 minutes, tied to something she said, and phrased in plain language without coaching jargon. Add one question about what she noticed when she tried the agreed experiment.
[paste your own summary, no names]

A sample output: "Which conversation this week did you handle differently from your usual approach, and what made that possible?" and "When you handed over the report, what did you tell yourself?" Edit to match your voice; keep the experiment question. A rule worth adopting: you write the session summary yourself, from your own notes, before using any tool. A summary generated by AI from a recording is a separate and more sensitive choice, covered in AI session notes and follow-ups for coaches.

Role-specific variations

The context block is the same; the prompt changes with the role. Three illustrative adaptations.

• A first-time CEO of a 30-person firm. Ask for hypotheses about the shift from doing to directing, and for questions about what they have stopped doing since taking the role. Add the board dynamic to the context.
• A senior engineer promoted to manager. Ask for the tensions between technical credibility and managerial authority, and for questions that surface what they miss about their old job.
• A finance director who must present to non-specialists. Ask for a structure for telling a story from numbers, and for questions about what the audience needs to decide, not what the director wants to say.

In each case ask for hypotheses with evidence, not answers, and for questions in your own style. If the model returns a list of tips, repeat the prompt with "do not give advice; give questions and hypotheses".

The same request, generic prompt versus context prompt

It is worth seeing the difference side by side, because the gap is the argument for the context block. Illustrative only.

Generic prompt: "Give me coaching questions for a leader who struggles to delegate." The output is a list of ten sound but interchangeable questions: "What are you afraid will happen if you delegate?", "What would your best day look like?", "What is the cost of holding on?". Any coach could have written them, and the client could have found them on the first search result. They are not wrong, but they are not about her.

Context prompt: the same request preceded by the context block. Now the questions refer to what is in the block: "You were promoted for being the person who fixes things. What does fixing give you that directing does not?", "Your team says it is hard to know what you want. If they could ask you one question on Monday morning, what would it be?", "You handed over the weekly report last time. What did you tell yourself when it came back slightly different from how you would have done it?" Each of these could only be asked of this client, and you could not have got them without the history and the stakeholder themes.

The lesson is that quality is mostly an input problem. If an output feels generic, add the missing context line (what has been tried, what the stakeholders said, what the client asked for today) rather than rewording the instruction.

Using ChatGPT for the three-way contracting meeting

Executive coaching often has a sponsor: a line manager or HR lead who pays and wants outcomes. The contracting meeting between coach, client and sponsor sets what success looks like and what is confidential. A prompt can help you prepare for it without involving any client material at all.

I am an executive coach preparing a three-way contracting meeting for a coaching engagement of six sessions over four months. The client is a newly promoted operations director; the sponsor is the CEO. Draft: (1) eight questions I should ask the sponsor about what they hope to see, (2) six questions for the client about what they want from coaching and what worries them, (3) a short list of confidentiality points I should make explicit, such as what I will and will not share with the sponsor, (4) three ways the sponsor's and the client's goals could conflict and how I might name that early.

This prompt carries no names and no confidential detail, since it describes roles only. A typical output includes "What would you need to see to say this was worth the investment?" for the sponsor and "What would you not want me to report back?" for the client. What you fix: the confidentiality list. You decide what you share and with whom, based on your own code and agreement; the model's list is a prompt for your thinking, not a standard. Write the final points yourself.

Closing an engagement: the review prompt

The last session is when a coach is most tempted to produce a neat summary. A review prompt helps you structure what you ask, while the evidence comes from the client.

I am preparing the final session of a six-session coaching engagement. Using the goal frame and the session summaries below (anonymised, written by me), suggest: (1) six questions that help the client name what has changed, with examples from their own experience, (2) three questions about what might slip once coaching ends and how they could prevent it, (3) a simple format for a one-page "what I will keep doing" plan the client writes herself.

The output is again questions and a format. Resist letting the model write the "what has changed" narrative. Clients remember what they say more than what they read, and a summary in the coach's words or a model's words tends to lack the detail that makes it theirs. If the sponsor requires a report, write the confirmed observable changes yourself and have the client approve what you share.

Measuring whether the prep is helping

Three signals, none of which need software. First, how often the session goes somewhere you had not planned and it is productive; good prep gives you options, not a script. Second, how often the client says "that is exactly it" or goes quiet after a question; specific questions land differently from generic ones. Third, your own prep time. If it has dropped but the quality of the questions has not, the tool is earning its place. If prep time has gone up because you are editing long outputs, tighten the prompts and use fewer of them.

Ask for client feedback occasionally: "Were the questions I brought today useful?" Not "did you like the AI?", which is a different question. Keep an eye on the client's comfort with how you work, and be ready to stop using the tool for their sessions if they ask.

Confidentiality rules to hold and to state

A coaching relationship rests on trust, and the rules about AI should be as deliberate as your rules about notes.

• Check what you have promised. Read your agreement, your professional body's code and any sponsor contract. If any of them restricts third-party tools, follow it.
• Anonymise at source. Describe roles, not people. No names, employers, locations, health or family details, and nothing from the client's confidential company information.
• Never paste raw feedback with names, recordings or transcripts into a consumer account. If you want transcription, that is a separate decision with its own consent and tool choice.
• Get informed consent. Put one clear paragraph in your agreement: which tools you use, for which purposes (preparation, not decisions about the client), what you never include, and that you review all outputs.
• Use projects and settings carefully. OpenAI says project instructions apply only within that project and override global custom instructions, so keep one project per client and keep the global instructions about your method, not about people. OpenAI also notes that information from your use of custom instructions can be used to improve model performance unless you opt out; check your data controls and keep client facts out of that field.
• Delete when finished. Remove working chats and keep your notes in your own secure system.

The consent paragraph might read: "I may use AI tools to help me prepare for sessions. I do not enter your name, your organisation's name or confidential details, and I do not use AI to make decisions about you. You can ask me not to use these tools for your work." Adapt it with your own legal or professional advice.

A worked session-prep example, with timings

Illustration: a coach preparing for a Thursday session with an operations director has 45 minutes on Wednesday evening. The time goes like this: 10 minutes updating the context block from last session's notes; 10 minutes on prompt 1 and reading the output critically; 10 minutes on the feedback themes (the interviews were done two weeks ago, so the synthesis already exists); 10 minutes choosing three questions and one structure and writing them in the coach's own words; 5 minutes drafting the reflection questions to send after the session.

The result is not a script. It is two hypotheses held lightly, four questions, one structure available if the client asks for it, and a plain awareness of the coach's own risk of over-leading. Without the tool, most of this happens anyway, but in the car, often incomplete. With it, the prep is written and checkable. What the model has not done is decide what the session is about. That is agreed with the client at the start, in the room.

Where this goes wrong, and how you notice

• Flattering hypotheses. The model tends to affirm the client's story. Add a line asking for a hypothesis that contradicts the client's own framing, and look at the evidence for each.
• Leading questions. "Don't you think..." slips in. Read every question aloud and rewrite any that contain the answer.
• Over-reading feedback. A theme from two of nine respondents becomes a "key issue". Keep the counts visible and say so to the client.
• Dependence. If you cannot prepare without the tool, you have swapped skill for convenience. Do some sessions from your own notes alone.
• Identifiable detail creeping back. A phrase like "the only woman on the executive team" identifies someone. Read your context block as if a stranger might see it.

For a wider view of how coaches keep their own voice when using AI, how business coaches use AI without losing the personal touch is a good companion, and whether coaching-specific AI tools are worth paying for covers when to buy something built for the job.

Building your own prompt library

After a few months you will notice which prompts you reuse. Keep them in a document with the context-block template at the top, and tag each prompt with the stage it supports: contracting, early sessions, feedback review, goal setting, wrap-up. Update them after sessions: if a prompt keeps producing leading questions, change the instruction. Store the document in your own files, not only inside a chat tool. If you would like help designing that library and the consent wording together, that is what my AI implementation consultation covers.

Questions executive coaches ask about ChatGPT

Is it safe to put a coaching client's details into ChatGPT?

Not without care. Coaching conversations are confidential, and your agreement or professional code may restrict third-party tools. Use anonymised role descriptions and placeholders, avoid names and company details, and get the client's informed consent for any use of AI that touches their material.

What makes ChatGPT session prep generic?

Missing context. If you give only a topic such as 'delegation', you get a textbook answer. Give the role, the situation, what has been tried, what the stakeholders said and what the client wants from the session, and the prep becomes specific.

Can ChatGPT summarise 360 feedback?

It can group and summarise themes if you strip names and identifying details first. Treat the summary as a draft: check it against the raw comments, watch for comments it blends together, and never let it add interpretation you would not defend in the room.

Should I tell clients I use AI in session preparation?

Yes, in plain words, and ideally in your agreement: what you use it for, what you never put in, and that you review everything. Many clients are comfortable with this when it is explained and unsettled when they discover it later.

Further reads

• How Coaches Use AI to Build Workbooks, Exercises and Programmes — Build workbooks and exercises with AI.
• ChatGPT for Coaching and Feedback: Session Summaries That Stay Personal — Session summaries and feedback that stay personal.
• Should Coaches Build an AI Version of Their Method for Clients? — Should you build an AI version of your method?
• How Coaches Use AI to Qualify Discovery Call Bookings — Qualify discovery calls with AI.
• What to Do When AI Gets Something Wrong With a Customer — What to do when AI output goes wrong with a client.
• AI Coaching for Leaders: What to Offer Clients Alongside Your Sessions — What AI coaching to offer leaders alongside your sessions.
• AI Note Taker for Executive Coaches: Keeping Sessions Confidential — Otter, Fireflies, Fathom, built-in and local options compared on training and retention, with settings to change and a client consent clause.
• AI 360 Feedback for Coaches: Collect, Theme and Report — A coach's 360 process from question set to debrief: anonymity thresholds, AI theme coding with a human check, a report clients trust, and what tools cost.
• AI Tools for a Coaching Business: Notes, Programmes and Follow-Up — A coach's practical workflow for consent, session records, programme templates and personal follow-up.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI Help Center pages on ChatGPT custom instructions and Projects. Prompts, notes and outputs in this tutorial are illustrative; no client material was used.