A coach can use an approved meeting tool for session summaries, a business AI workspace for de-identified programme templates, and their existing client system for reviewed follow-up drafts. First obtain informed consent for any recording or AI processing, check the supplier's data terms, restrict access, and keep the coach responsible for interpreting and sending each note.
Do not treat confidentiality as a setting you switch on once. A coaching session can reveal work concerns, family circumstances and sensitive goals in one conversation. The International Coaching Federation's Code of Ethics, in its current version since 1 April 2025, calls for a clear agreement about what information is exchanged, records kept and disposed of confidentially, and ethical duties met through any technology a coach uses, artificial intelligence included. A tool's “no training” statement answers only one part of that duty; sharing, retention, access and client expectations still need decisions.
Separate the three jobs before buying a tool
Session notes, programme design and follow-up have different inputs. Notes begin with a private conversation and need consent, accuracy checks and retention rules. Programme design can often begin from a reusable, anonymous learning objective. Follow-up needs only the agreed action, date and tone, not a full transcript. If one tool handles all three, configure each job separately; if it cannot restrict what each job sees, use a narrower workflow.
Zoom's support article lists Meeting Summary with AI for licensed users on Zoom Workplace Pro and higher plans. Only the host or co-host can start or stop it during a meeting, a host can set it to start automatically, participants see an indicator while AI features are active, and summaries can be emailed to participants if the host chooses to share them. It works from speech-to-text, so it runs whether or not the meeting is recorded. That makes it a candidate for a coach already meeting there, subject to the coach's contract and local recording obligations. Confirm exactly what is generated, who receives it and whether a participant can decline. A coach who promises a private conversation should not discover after the call that a summary was automatically shared to the wrong audience.
ChatGPT Business is one possible workspace for templates and approved drafts. OpenAI states that it does not train its models on ChatGPT Business data by default. That does not grant permission to upload client material or settle retention and access questions. Review the coaching agreement and the supplier terms first. For a deeper note-taking product decision, see the executive-coach note-taker confidentiality checklist.
| Job | Minimum input | Required human decision |
|---|---|---|
| Session note | Consented recording or coach's own brief notes | What is accurate, necessary and appropriate to retain? |
| Programme outline | Anonymous goal, session count and exercise constraints | Does the sequence fit this client's agreed goals? |
| Follow-up | Confirmed action, owner and due date | Does the message represent what the client agreed? |
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowAgree consent and a no-recording route
Explain before the session whether audio is recorded, whether AI creates a summary, which supplier processes it, who may receive it, how long records are kept and how a client can choose a session without that process. Put the choice in the coaching agreement and confirm it again when the feature is activated. A meeting platform's notice is helpful, but it does not replace a conversation about the coaching relationship. Ask a qualified adviser about legal consent and retention rules that apply to your practice.
Sample wording to adapt: “With your agreement, I can use [approved tool] to create a draft session summary. I will review it before storing or sharing it. We can work without recording or AI notes; I will write brief notes myself. Please tell me if you want that option.” This is illustrative wording, not a universal consent clause. Match it to what the tool actually does and what your contract says. Never offer an opt-out that the platform cannot honour.
Record the choice as a field in the client record, with date and scope. “Consented to session 3 recording” is different from “consented to every future session”. A coach might use AI for a goal-setting call and then switch it off for a more sensitive conversation. If a client changes their mind mid-session, stop the feature and follow the agreed handling procedure for material already captured. Test that procedure on a dummy meeting before relying on it.
A realistic failure (illustrative): the coach disables recording but leaves meeting summary active, assuming they are the same control. The platform still creates a text summary. The fix is to test recording and summary settings separately, check any participant notification, and inspect where the output lands after a dummy call. A checklist that says merely “recording off” would miss this distinction.
Write a consent register that distinguishes purposes. The columns can be “client code”, “session date”, “audio recording”, “AI summary”, “summary shared with client”, “sponsor access”, “withdrawal request” and “coach confirmation”. Filled in with illustrative entries, one row might say: “C04; 6 November; no; no; coach's manual action note only; none; not applicable; confirmed before start.” Another might allow an AI summary for one session while denying sponsor access. This prevents a single generic consent tick from being interpreted as permission for every future use of the material.
For a sponsored engagement, use an especially clear boundary. A manager who pays for coaching may ask for “a quick update” after the session. If the agreement permits only attendance and progress against broad goals, do not use AI to produce a detailed session recap for that manager. An illustrative appropriate update might be “Two of six sessions completed; next review date agreed.” It would omit the client's private examples, concerns and tentative career plans. The coach should check the agreed information-sharing rule before writing any sponsor message.
Make a session note from the coach's evidence
Decide what a useful note contains: agreed goal, client-stated insight, action, deadline, matters to revisit and any correction the client requests. Do not build a transcript archive merely because a tool can make one. A shorter record may reduce unnecessary exposure and make future sessions easier to prepare. The coach should distinguish a client's words from the model's interpretation, especially around motivation, emotion or relationships.
After a consented session, review the draft while the conversation is fresh. Check names, dates, negations, commitments and whether the summary attributes an idea to the right person. Remove speculative labels such as “resistant to change” unless the coach has a sound, agreed reason to record that interpretation. Mark uncertainty plainly. If the tool says “client will resign next month” when the client said “I might consider a change”, correct it before the note enters the client record.
Before and after, with illustrative wording. AI draft: “Client committed to speaking to their manager on Friday.” Coach's checked note: “Client will consider whether to request a meeting with their manager; will decide by Friday.” The first version turns a decision date into a promised conversation. That would make the next session awkward and could create pressure the client did not accept. The correction is small in words and large in meaning.
If the client chose no recording, use the coach's own brief notes. A de-identified prompt could turn “goal: delegate weekly rota; obstacle: unclear handover; action: draft checklist before next session” into a tidy private note. Remove names, employer details and any unusual fact that could re-identify the client. If de-identification would strip away the context needed for a useful summary, write the note manually in the approved client system. The weekly notes and follow-up workflow gives the narrower operational steps.
Design programmes from anonymous learning goals
Programme design is usually the safest of the three jobs to pilot because it can begin with a generic objective. Ask an AI assistant to propose a sequence of sessions, practice exercises and reflection questions for a type of goal, not a named client's intimate history. Then compare the outline with your coaching method, the actual agreement and the client's preferences. An AI plan is a draft of activities, not an assessment of a person.
For example, an illustrative six-session programme for a manager practising delegation might move from defining outcomes, to mapping current decisions, to drafting a handover, to rehearsing a conversation, to reviewing one real attempt, to deciding what to maintain. Ask for two exercise options per session, one that can be done in ten minutes and one that takes half an hour. The coach chooses based on the client's capacity. Avoid a rigid “week 3 must do X” structure if the work uncovers a different priority.
A copyable planning prompt is below. Its inputs are intentionally general. The coach should replace the bracketed fields with an anonymous goal, not a life story. Ask for assumptions and a reason for each exercise so that weak suggestions are visible.
Draft a six-session coaching programme for a client working on
[general goal]. Each session is [duration]. Offer one reflection
question, one practice exercise, and one way to review progress.
Use only the information supplied. Do not diagnose the client,
invent history, or promise an outcome. Mark assumptions.An illustrative output might suggest “send a daily progress score to the coach”. The coach rejects it because the agreement covers sessions and one weekly check-in, not daily monitoring. A better exercise is a private decision log the client can choose to discuss. AI proposed an activity; the contract and coaching judgement determined whether it belonged. For a workbook-focused workflow, use the exercises and programme tutorial.
To check programme quality, ask the assistant for the purpose and possible burden of each exercise. Suppose session two proposes a “360-degree survey” but the client has no authority to request feedback from colleagues and the programme has no consent process for respondents. Replace it with a self-observation exercise: write down three decisions delegated during the week and what information the recipient needed. That keeps the learning goal while removing an unnecessary data collection step. Do not accept a named method merely because it sounds professional.
Keep versions of the programme. The first outline might allocate a whole session to a goal the client decides is no longer relevant. Mark the change in the coach's record and ask AI, using only an anonymous summary of the new goal, for two alternative exercises. Compare them with the client's preferences. The programme should adapt to the coaching conversation rather than force the conversation to match an AI-generated sequence. A revised template is a tool for preparation, not a promise of a fixed result.
Draft follow-up from agreed actions only
Use a compact action record: “Action”, “Owner”, “Due or review date”, “Client wording”, “Shareable by email?” and “Coach check”. Generate a message from that record, not from the full transcript. The first review asks whether the action really was agreed. The second asks whether the message is suitable for the channel: email may be forwarded or seen by another person. Avoid including sensitive reflections in a subject line or notification preview.
An illustrative record says: “Client will draft a two-column list of tasks to keep and delegate by 18 November; private reflection about manager excluded from email.” AI draft: “Dear [first name], thank you for today's session. As agreed, please bring your keep/delegate list to our next meeting on 18 November. Let me know if the date needs changing.” The coach checks that 18 November is the meeting date, not the deadline for completing the list, then adjusts the sentence if needed.
A follow-up can also be a choice rather than a reminder. Suppose the client has not agreed to share an exercise. A safe draft might say, “If you would like to discuss your notes next time, bring the parts you want to explore.” It should not say, “Please send me your completed worksheet tonight.” This distinction protects client agency and avoids creating a new information flow the agreement never covered. AI is especially prone to turning a tentative suggestion into a firm task because decisive emails read smoothly.
Check delivery settings. A CRM automation that sends a draft immediately after a call bypasses the coach's review even if the wording is good most of the time. Start with draft-only status and a named approver. Test a dummy client whose email address deliberately differs by one character from another dummy record; the reviewer must catch the recipient as well as the text. Sending a private coaching summary to the wrong person is a serious failure, not a minor formatting defect.
Build a short “do not include” list for follow-up drafts: sensitive disclosures, third-party names, unagreed interpretations, diagnosis-like labels and any action still under consideration. Suppose a draft (illustrative) says “You recognised that your colleague is undermining you.” The session note only records that the client is uncertain about a difficult working relationship. The coach removes the assertion and, if useful, writes “You wanted time to reflect on the conversation before choosing a next step.” The second sentence stays close to what was agreed and is less likely to cause harm if forwarded.
A solo coach's month of 24 sessions, timed and audited
Put numbers on it: a solo coach (figures illustrative) has 12 clients, each with two sessions a month, so 24 sessions. The coach currently spends 15 minutes per session on notes and 10 minutes on follow-up, or 25 × 24 = 600 minutes, ten hours. They pilot AI notes only with four clients who give informed consent, making eight sessions. Programme templates are built separately from anonymous goals.
For those eight sessions, generation plus checking takes 9 minutes for each note and 8 minutes for each follow-up draft and send, or 17 × 8 = 136 minutes. The previous manual process would have taken 200 minutes for the same eight, so the illustrative difference is 64 minutes. But setup, consent administration and supplier review might take several hours initially. The coach should not present a one-month net saving without counting those tasks or any correction time.
Use four acceptance tests: every recorded session has documented consent; every note correctly distinguishes an intention from a commitment; every follow-up has the right recipient and agreed action; and the client can choose a no-AI route without losing the service. In the sample pilot, one note wrongly states a firm commitment. The coach corrects it, logs the pattern and adds an explicit “tentative versus agreed” check before expanding the pilot. The timing result remains useful, but quality determines whether the process continues.
A second illustrative measure is the time to locate an action at the next session. If the old notes take three minutes to search and the new structured notes take one, the benefit is not only drafting time. However, the structured note must preserve the client's meaning. Track both retrieval time and correction count; a fast-to-find wrong action is worse than a slower accurate one. The coaching package audit helps decide which parts of the service should remain fully human.
Review the ethical fit before scaling
Return to the ICF Code of Ethics when writing your operating rule. It calls for clear agreements, confidentiality, proper handling of records and responsibility for technology used in coaching. Write down who can access notes, what is retained, how errors are corrected, what happens if a client withdraws consent, and when AI must be off. If a sponsor pays for coaching, the agreement must also specify what, if anything, the sponsor receives; do not assume a summary can be shared because the sponsor funds the work.
Keep a small audit each month. Select two generated notes and two follow-ups, trace each action to what was actually agreed, check access permissions and verify the recipient. Ask clients whether the notes feel accurate and useful, without pressuring them to accept the technology. If they prefer a handwritten summary, honour the agreed route. A model can help organise work, but the trust, judgement and confidentiality of the coaching relationship remain the coach's responsibility.
Further reads
- Best AI Tool for Coaching: Notes, Scheduling and Content Compared — Compare scheduling, notes and content products by business job.
- How Coaches Use AI to Qualify Discovery Call Bookings — Handle initial enquiries before any confidential session begins.
- ChatGPT Prompts for Executive Coaching: Prep, Reflection and Goals — Prepare prompts around agreed goals rather than generic advice.
- ChatGPT for Coaching and Feedback: Session Summaries That Stay Personal — Review client feedback and summary language in detail.
- Best AI Model for Life Coaching: What Session Notes and Privacy Need — Compare model choices after setting your confidentiality boundary.
- 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.
- How to Use AI to Manage Projects Across Several Clients — One task board, a weekly AI status digest and deadline alerts for a small firm running six client projects at once, with the setup steps and sample outputs.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: International Coaching Federation, ICF Code of Ethics (in effect from 1 April 2025); Zoom Support, Using Meeting Summary with AI; OpenAI, Business data privacy, security and compliance (all checked 4 November 2026).