Best AI Model for Life Coaching: What Session Notes and Privacy Need

For session prep and follow-ups, any of ChatGPT, Claude or Gemini can do the job, so choose by privacy plan and by whose drafts sound most like you. Use a paid or business plan with training switched off, give the model only anonymised summaries, and never paste names, health details, trauma disclosures or anything a client said in confidence.

That answer disappoints people who want a winner. The honest reason is that coaching work asks little of a model's raw ability. Turning a bullet summary into a warm, clear follow-up email is something all three do well, and they all change every few months. What does not change is the duty you carry to the client, and it is that duty, not a benchmark, that should drive the choice.

Follow me on Instagram@sagnikteaches

What a coach actually needs from a model

Write the job list before you compare products. For most life coaches it is short.

Connect on LinkedInSagnik Bhattacharya

• Session prep. Turning your anonymised notes from the last session into a one-page prep sheet: themes, the client's stated goals, unresolved questions, and two or three possible reflective questions.
• Follow-ups. Drafting a short email after each session that recaps the client's own commitments in their words.
• Exercises and resources. Producing a worksheet, a reflection prompt or a reading list for a topic such as values or boundaries.
• Admin. Reminders, scheduling emails, invoices wording, social posts.

None of that needs a frontier reasoning model. It needs a model that follows instructions, keeps a consistent tone, does not invent facts, and sits inside settings you can defend to a client. Notice what is not on the list: assessing a client's mental state, giving health advice, or deciding what a client should do. That boundary is covered in AI for life coaches between sessions within the ICF Code.

Subscribe on YouTube@sagnikteaches

The ICF Code: what it says that bears on AI

The ICF Code of Ethics is the standard many coaches work to. I checked the current text, in force since 1 April 2025. The standards that matter sit in its confidentiality section, quoted briefly here and then paraphrased.

• 2.1 asks you to "maintain the strictest level of confidentiality with all parties involved". Pasting a client's story into a public tool is a confidentiality question before it is a technology one.
• 2.2 asks for a clear agreement about what information is exchanged and how it is exchanged among all parties involved. This is where your coaching agreement should mention AI.
• 2.3 asks for a clear agreement with clients, sponsors and other parties about what confidential information may need to be disclosed to the appropriate authorities, for example where there is imminent or likely risk of danger to self or others. That agreement, not a chatbot, is what you follow if a client raises risk.
• 2.4 asks you to maintain, store and dispose of records, including electronic files and communications, in a way that promotes confidentiality, security and privacy and complies with applicable laws and agreements. Notes stored in a chatbot's history are records.
• 2.5 says you fulfil your ethical and legal obligations directly and through any technology systems you use, and its list of those systems names artificial intelligence. The appendix defines AI as any algorithm or machine-based technology that enables computers and other digital devices to simulate human intelligence. In plain words: the tool is yours to answer for.

The Code does not ban AI. It says the obligations follow you into every system you use. That is the test to apply to each model and plan: could I explain to this client, calmly, exactly where their information went?

What never goes into any model

Set this as a hard list and stick it on the wall by your screen. Whatever model you choose, these stay out, even on a paid plan.

• The client's full name, email, phone number, address, employer or any combination that could identify them. Use a code such as "Client M" or a first-name-free label.
• Health information, mental health history, medications, diagnoses and anything the client mentions about therapy or treatment.
• Details of trauma, abuse, relationship breakdowns, legal trouble or finances that identify the person or others involved.
• Names of third parties in the client's life: partners, children, colleagues.
• Recordings, full transcripts and uploads of session audio.
• Anything the client asked you to keep between you, and anything covered by a confidentiality clause with a sponsor, such as an employer paying for coaching.

What can go in: your own anonymised summary in your own words, the themes without the story, and the client's stated goals in general terms. Anonymising is more than deleting a name. A summary that says "a senior engineer at a small bakery chain who is going through a divorce" may still identify someone to anyone who knows them. Err towards vagueness.

Privacy settings: what to check on each plan

Privacy settings differ between consumer and business plans, and labels change, so confirm each one in your own account. The points below were checked on the vendors' own pages in November 2026.

Two practical conclusions. First, a sole coach who wants a business plan with training off by default may have to pay for two seats on ChatGPT or Claude, roughly $40 to $50 a month, or choose a personal paid plan and switch training off by hand. Second, if you already pay for Google Workspace, Gemini inside it may be the cheapest way to get a business-grade privacy position. In all cases, read the vendor's current terms, because memory features and defaults change.

One anonymised session run through three models

Here is a method you can repeat yourself, with an illustrative summary. I am not reporting a benchmark: models change monthly, so the useful thing is the test you run, not my ranking of them.

The anonymised summary, written by the coach in her own words after a session:

Client K, 3rd session, goal: "stop saying yes to everything".
Themes: guilt when declining; a pattern of overcommitting at work
and in family plans; energy low by Thursday.
Client's own commitments: say "let me check and get back to you"
once this week; block Thursday evening for herself.
Open questions: what is she afraid will happen if she says no?
Next session: review how the "check and get back" experiment went.

The prompt, identical for each model:

You are helping a life coach write a follow-up email. Use only the
summary below. Do not add facts, diagnoses or advice. Keep it under
150 words, warm and plain, in the first person as the coach. Recap
the client's own commitments in her words and end with one open
question. Do not use the word "journey".

[paste the summary above]

Score each reply against a five-line rubric, one point per line:

1. Uses only what was in the summary (no invented detail).
2. Under 150 words and no filler.
3. Recaps the client's commitments as hers, not as instructions.
4. Sounds like you, not like a template.
5. Contains no advice, no diagnosis, no therapeutic language.

An illustrative reply, of the kind any of the three might give, and what you would fix in it:

Hi [first name],

Thank you for today. I liked how honest you were about how hard
it feels to say no. This week you're going to try "let me check
and get back to you" once, and to keep Thursday evening for
yourself. That's a small, real step. It might also be worth
noticing what comes up in your body when you pause before answering.

What are you most afraid might happen if you say no?

[your name]

Rubric marks: it keeps to the summary and the length, recaps the commitments, and ends with the open question. The fix: "what comes up in your body" is an addition that was not in the summary and drifts towards therapeutic territory, so cut it. Also "I liked how honest you were" is a judgement the coach may not wish to make. After trimming, the email is the coach's own. Run the same prompt on each model, score them, and note which one needs the fewest edits. That is your answer for now, and you can repeat the test each quarter in twenty minutes.

How the three tend to differ in use

These are tendencies you should check against your own test, not rankings, since each vendor updates models often and I have not run controlled comparisons.

• ChatGPT. Widely used, with Projects for keeping standing instructions such as your tone guide and a "no advice" rule. If memory is on, details can carry from one chat to the next; Temporary Chats do not create or update memories. Check what is remembered and switch memory off for client work if you prefer.
• Claude. Projects are available on every plan including free (free accounts can create up to five), which suits keeping a tone guide and templates in one place. Judge its drafts against your own voice, exactly as you would the other two.
• Gemini. Convenient if you live in Gmail, Docs and Calendar, particularly with Workspace, where it sits inside tools you already use. Gems, its reusable instructions, are becoming "skills": Google's help page says the switch starts in November 2026 for personal accounts, with Workspace business accounts following in March 2027.

If all three score the same on your rubric, pick the one with the cleanest privacy position for your situation and stop comparing. Switching later costs you only the standing instructions, which you can paste into any of them.

Setting up a safe routine

1. Choose the plan first. Paid or business, training off, history policy decided. Write down the setting names and the date you checked.
2. Write standing instructions. Tone, length, banned words, "use only what I give you", "no advice or diagnosis". Keep them in a Project, Gem or skill, not retyped each time.
3. Make anonymising a habit. Write your notes with codes from the start so there is no identifying detail to remove later. Better still, write the summary on paper or in your own notes app and paste only that.
4. Draft, then edit. Never send AI text without reading it aloud. Remove anything you would not say.
5. Delete when done. Delete the conversation or use a temporary chat after copying the final text, in line with your retention policy.
6. Keep a one-line log. Date, task, plan, anonymised yes. It makes a client question or a supervision conversation easy to answer.

Wording for your coaching agreement

Sample text to adapt, not legal advice. It covers disclosure and consent in a way that supports the ICF Code's standard on agreeing how information is exchanged.

Use of AI tools. I may use AI writing tools to help me draft
administrative messages, such as session follow-up emails and
worksheets. I do not enter your name, contact details, health
information or the details of what you share in sessions. I enter
only brief, anonymised summaries in my own words, on accounts with
data-training switched off, and I review everything before it is
sent. AI is not used to assess you or to make coaching decisions.
If you would prefer I did not use AI for your sessions, tell me
and I will not.

The opt-out line costs you almost nothing and is worth having. A client who does use it will be glad to have been asked, and you will have a clear record of the choice.

Five coaching situations that test your AI rules

A client sends a long, emotional message between sessions

The temptation is to paste it in and ask for a reply. Do not. It contains identifying and sensitive detail, and the right response is yours, perhaps shorter and kinder than a model's. If you want help, describe the situation in neutral terms yourself ("client is upset about a family event; I want to acknowledge and keep the next session on track") and ask for options on tone only.

A corporate sponsor pays for the coaching

Some sponsor contracts restrict what can be shared with third parties, and the ICF Code's Standard 2.3 expects a clear agreement with sponsors about confidential information. Treat AI vendors as third parties unless the contract says otherwise, and check before using any tool for that client's work.

You record sessions to take notes

Recording and automated transcription add a layer of risk beyond summaries: a full transcript is rich with identifying detail. If you use a note taker at all, get explicit consent, read its retention terms, and see AI note takers and confidentiality for coaches before deciding.

A client asks you to analyse their personality or "what's wrong"

Do not run it through a model. Assessments of this kind move into diagnostic territory, and your role is coaching. If the client needs a clinical view, refer them to a qualified professional and say so in plain words.

You run group programmes with many participants

Group data multiplies the risk. Use aggregate, anonymous themes only, for example "most participants named time as their main obstacle", and keep participant lists, names and comments out of any model.

Hours saved across 24 sessions a month

Illustrative figures. A coach with 12 active clients, seeing each every two weeks, has about 24 sessions a month. Without AI she spends about 15 minutes per session on prep and 10 on the follow-up email, around 10 hours a month. With an anonymised-summary workflow, prep drops to about 10 minutes and the email to about 5 once the standing instructions are set, around 6 hours. The saving is about 4 hours a month. On a $20 plan that is $5 an hour of cost for time she can spend on clients or on rest, a worthwhile trade only if the safeguards above are in place. If she is spending 20 minutes per session on anonymising and checking, the saving shrinks, which is a useful signal to adjust the routine rather than push on. For how this fits into the wider picture of repricing, life coaching rates during automation covers it.

Tightening a standing instruction that breeds clichés

Most poor AI drafts trace back to loose standing instructions. Here is a realistic first attempt and its repair, for a coach who keeps getting emails full of motivational clichés.

BEFORE
Write encouraging follow-up emails for my coaching clients.

AFTER
Write follow-up emails from me, a life coach, to a client.
Use only the summary I paste. Never add facts, advice, diagnoses
or reasons for how the client feels. Plain British English, under
150 words, no exclamation marks, and none of these words: journey,
empower, unlock, transform. Quote the client's own commitments.
End with one open question taken from my "open questions" line.
If the summary lacks something, write "not provided".

The second version names what to leave out, which matters more than what to include. It also gives the model a way to admit a gap instead of filling it. Keep this text in a Project, Gem, skill or saved note, and update it whenever you edit a draft in the same way twice.

A quick check that anonymising worked

Before pasting any summary, give it a thirty-second test. Read it as if you were the client's friend, manager or relative. Could they tell who it is about? Count the specific facts (job, place, family situation, dates, unusual events). If more than two are still present, remove or blur them. Then ask whether the model needs each remaining fact to do the task. For a follow-up email it almost never needs the job or the town. A good rule is to keep only the themes and the client's own commitments, because those are what the email actually uses.

Privacy slips coaches make without noticing

• Assuming "private mode" is private. Temporary or incognito chats still pass through the vendor's systems and are kept for a period. They help; they are not a licence to paste identifying details.
• Letting memory build a client profile. If memory is on, the tool may carry details across conversations. Turn it off or review it for client work.
• Using the free tier by accident. A personal free account may have different training defaults from a paid plan. Check which account is signed in before pasting anything.
• Sending unedited AI text. It sounds slightly off, and clients notice. Always edit.
• Letting the tool coach. Asking a model what the client should do, then relaying it, stops being your coaching. Keep the questions and the judgement yours.

Choosing today, and when to choose again

If you are on a business plan with training off, run the three-model test above and choose the one needing fewest edits. If you are on a free consumer account, move to a paid plan or switch training off before any client-related use. If you are unsure of your settings, use none of the tools for client work until you have checked. If a client opts out, do the drafting by hand for that client. Review your choice each quarter, because plans, defaults and models move quickly.

For the wider toolkit, the best AI tool for coaching businesses compares notes, scheduling and content tools, and AI session notes and follow-ups for coaches gives a weekly workflow. The AI guide links the rest. If you want help setting up the settings, prompts and agreement wording for your own practice, that is what my AI implementation consultation covers, through the contact page. None of this is medical, therapeutic or legal advice.

Questions life coaches ask about choosing an AI model

Is one model clearly the best for coaching?

No model wins on every task, and models change every few months. For coaching work the plan and its privacy terms matter more than the model name. Pick the one whose settings you can verify and whose follow-up drafts sound like you, then test it on a made-up client.

Can I paste a real client's session notes into ChatGPT?

Not as raw notes. Names, identifying details, health information and anything the client said in confidence should stay out. Use an anonymised summary on a plan where training is off, and check your client agreement and the ICF Code's confidentiality standards first.

Do I need to tell clients I use AI?

It is good practice and your confidentiality agreement may require it. Say in your contract which tasks AI helps with, such as drafting follow-up emails from anonymised notes, and that you do not enter identifying details. Offer clients the option to opt out.

Can AI replace my coaching judgement or help with a client in crisis?

No. Use AI for admin and drafting only. It is not a therapist and should not be presented as one. If a client raises risk to themselves or others, follow your agreed disclosure and referral steps, not a chatbot's suggestions.

Further reads

• Life Coach Questionnaire for Clients: Build and Analyse It With AI — Building and summarising an intake questionnaire safely.
• AI Tools for a Coaching Business: Notes, Programmes and Follow-Up — Tools for notes, programmes and follow-up in a coaching business.
• Are AI Tools Built for Coaches Worth Paying For Over ChatGPT? — Whether coach-specific AI tools beat a general chatbot.
• How Business Coaches Use AI Without Losing the Personal Touch — Keeping your own voice when AI drafts the admin.
• Can Therapists Use ChatGPT for Session Notes? — Why session notes are a harder case in therapy.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the ICF Code of Ethics (in force from 1 April 2025, Standards 2.1 to 2.5 and its appendix); OpenAI's ChatGPT and Business pricing pages and Data Controls FAQ; Claude's pricing page, Anthropic's consumer-terms update and privacy centre; Google's Gemini subscriptions page, Gemini Apps privacy help, Workspace privacy hub and Gems-to-skills help page; all checked November 2026. Settings labels change, so confirm them in your own account.