AI for Life Coaches: Between-Session Support Within the ICF Code

A life coach can use AI between sessions for scheduling reminders, reflection prompts and homework recaps, provided nothing identifying goes in, the client has agreed in writing, and you review every message before it is sent. Under the ICF Code of Ethics, confidentiality and AI use are your responsibility, not the tool's.

That sentence hides most of the work. Coaches who get into trouble rarely do so by drafting a reminder with a chatbot. They get into trouble by pasting a session transcript into a free account, or by letting a tool message clients without a human reading it first. The safe jobs and the risky ones sit on either side of a handful of specific standards in the Code, and the wording and checklist below put that line into practice this week.

Follow me on Instagram@sagnikteaches

What the ICF Code actually says about technology and AI

The ICF Code of Ethics currently in force went into effect on 1 April 2025. Its definitions section now includes a definition of artificial intelligence, and Standard 2.5 names it directly: you fulfil your ethical and legal obligations to clients, sponsors, colleagues and the public both directly and through any technology you use, and the Code's list of that technology runs from coaching tools, databases and platforms to software and artificial intelligence. Read that twice. The obligation does not move to the vendor when you adopt a tool. If a chatbot stores something it should not, the breach is yours.

Connect on LinkedInSagnik Bhattacharya

Four other standards do the practical work for between-session support. You can check each against the full text on the ICF Code of Ethics page, because wording matters and I am paraphrasing here.

Subscribe on YouTube@sagnikteaches

• Standard 1.1 asks you to explain the nature of coaching and co-create an agreement before coaching begins, covering roles, responsibilities, confidentiality and other aspects of the engagement. This is where AI use is disclosed.
• Standard 2.1 requires the strictest level of confidentiality with all parties.
• Standard 2.2 asks for a clear agreement about what information is exchanged and how, which covers which channels carry reminders and what they may contain.
• Standard 2.4 concerns how you maintain, store and dispose of records, including electronic files and communications, in a way that promotes confidentiality, security and privacy.
• Standard 2.6 makes you responsible for the adherence of your support personnel. The Code defines support personnel as people who work for you and have access to client information, for example assistants or bookkeepers. An AI tool is not a person, but treating it like a new assistant is a sound test: would you let a stranger on their first day read this?

There is also a separate document, the ICF Artificial Intelligence (AI) Coaching Framework and Standards. ICF's resource page dates it 8 November 2024, and the current PDF is marked V1.01, revised 8 October 2024. It sets standards across six domains: foundational ethics, co-creating the relationship, effective communication, learning and growth facilitation, assurance and testing, and technical factors such as privacy and accessibility. ICF describes it as a guide for providers, developers and organisations implementing AI-driven coaching. If you are a solo coach drafting reminders, it is not your rulebook, but it is a good yardstick when you evaluate a product that claims to coach your clients between sessions. ICF also publishes a companion document, Questions Stakeholders Should Ask, written for people weighing up an AI coaching product.

A three-column sort: safe, conditional, never

Before any tool choice, sort every between-session job into one of three groups. The sort is what keeps you inside Standard 2.1, so do it on paper once and keep it with your coaching agreement.

The conditional column is where most coaches live, and it depends on settings you control. On personal ChatGPT accounts (Free, Plus and Pro), OpenAI's help centre says data sharing for training is on by default until you turn off "Improve the model for everyone", and other consumer apps have their own model-improvement setting. ChatGPT Business and Claude Team do not train on business content by default, according to both vendors' pricing pages in November 2026, but settings change, so confirm on the vendor's own page. For the how-to, see how to stop AI tools training on your business data. A one-person practice pays for two seats on ChatGPT Business or Claude Team, which are $20 per seat a month billed annually ($25 monthly) at list price, so many solo coaches use an individual paid plan with training switched off instead. Either way, an account with the training setting verified is a precondition, not an upgrade.

Job one: reminders that carry no content

The lowest-risk, highest-return job is the reminder. A scheduling tool can send "Your session is tomorrow at 10:00, here is the link", and AI can draft the template wording once. Nothing about the client's life belongs in it. Standard 2.2 is the reason: an email or text can be read over a shoulder, forwarded, or sit on a shared family phone, so the channel is a confidentiality decision.

Here is a prompt you can use to draft the template, and notice that it contains no client information at all.

Write three versions of a session reminder for a life coach. Each is under 40 words,
warm but plain, with placeholders [first name], [day], [time] and [link].
It must not mention goals, homework, feelings or anything about the coaching
content. Add one line that lets the client reschedule by replying "move".

A typical output (illustrative) reads: "Hi [first name], a reminder that we meet on [day] at [time]. Your link is [link]. If you need to shift it, reply 'move' and I will send options." What you would fix: the first draft often adds "Looking forward to hearing how your week went", which nudges the client to rehearse private matters in a message thread. Delete lines like that. A reminder that says nothing leaks nothing.

The test for any reminder is the lock-screen test. Imagine the notification appears on a phone left on a café table. If it embarrasses the client, it fails.

Job two: reflection prompts without personal data

Between sessions, many coaches send one question to keep a client thinking. AI is good at producing question banks, and the bank itself contains no client data, so this is a safe job when you keep it generic. Ask for prompts by theme rather than by person.

I am a life coach. Give me 12 reflection questions for the week between sessions on
the theme "making a decision I keep postponing". Each should be open, non-leading,
under 25 words, and must not assume a diagnosis, a relationship status or a job.
Group them by: noticing, weighing options, first small step.

Illustrative output: "What is the decision you most often rehearse and rarely make?" "If the outcome were guaranteed to be fine, which option would you pick?" "What is the smallest version of this step you could take in ten minutes?" Then you edit. Look for three flaws: leading questions ("Don't you think you should...?"), questions that presume a context ("When did your partner last...?"), and anything that edges into therapy territory. Coaching and therapy differ in scope, and a prompt that explores past trauma between sessions, with no one present, is a poor fit for either.

The conditional version personalises one prompt from the client's own stated goal, using wording the client has already written to you in an email or form. Keep the input to one sentence and strip names: "Client wants to change career within a year and keeps delaying updating a CV" is enough. Ask for three prompt options, pick one, and rewrite it in your voice before sending. If the client's goal reveals a diagnosis, a named employer or a family situation, generalise it further or do it by hand.

For a wider look at how coaches keep a personal tone while automating routine parts, how business coaches use AI without losing the personal touch is a useful companion.

Job three: homework recaps from your own notes

The homework recap is the job where the ICF checks matter most, because it is the one where session content is closest to the tool. The safe way is to reverse the usual flow. Do not give AI the session and ask it to summarise. Give AI your own three bullet points, written after the session in your words, and ask it to turn them into a short, friendly recap with the agreed action.

A worked example, with illustrative numbers. Say a life coach has 14 active clients on fortnightly sessions, so 28 sessions a month. Recaps written from scratch take about 12 minutes each, around 5 hours 36 minutes a month. Using the bullet-point method, writing the bullets takes 2 minutes, the tool drafts in seconds, and review and edit take 3 minutes. That is about 5 minutes a recap, 2 hours 20 minutes a month, so roughly 3 hours 16 minutes saved. These figures are my own estimates for illustration, not measurements, but they show the shape: the saving comes from the drafting, not from skipping the review.

The before and after below shows the input and the draft. The input contains no name and no identifying detail.

Notes (mine): agreed to try a 20-minute "worry window" at 5pm daily.
Client will track how often evening worry returns after it.
Wants to talk about boundaries with a relative next time.

Task: write a 90-word recap email to the client. Warm, plain English, no advice
beyond what is in my notes. Put the action in bold. Placeholder: [first name].

The draft (illustrative) comes back with a greeting, the 5pm worry window as a bold action, a line about tracking, and a closing sentence. What I would fix: the model almost always adds encouragement that sounds like advice ("This technique is proven to reduce anxiety"). That is a claim you cannot support and a drift towards therapy, so delete it. Next, the line about "boundaries with a relative" reveals a family matter in an email. Move it out of the recap and keep it for the session. The recap should carry the action and nothing else. A rule of thumb: if the message would be awkward read aloud in front of the client's colleagues, it has too much in it.

Add the client's own words back by hand

The last step is yours. Open the draft, replace "[first name]", and add one sentence that only you could write, such as a phrase the client used. That sentence is also the proof to yourself that you read the message. Coaches who skip it start sending near-identical recaps to everyone, which clients notice within a month.

The consent clause: where the Code asks you to say it out loud

Standard 1.1 puts the explanation before the first session, in the agreement. Nothing in the Code lists the exact words, so the clause below is a starting point to adapt, and a lawyer or your own supervisor can sharpen it. The aim is plain English that a client can say yes or no to.

Use of AI tools. I may use an AI writing assistant to draft scheduling reminders,
generic reflection questions and summaries of action points that I have written
myself. I do not enter your name, session recordings, transcripts or any detail you
share in confidence. I read and edit every message before it is sent. Tell me
at any time if you would prefer I do not use these tools for your work together,
and I will stop.

Four design choices in that clause are deliberate. It names the jobs, so the client knows the scope. It lists what is excluded, which does more to reassure clients than any promise. It states human review, because the Code holds you responsible for the output. And it gives an exit with no penalty, because Standard 1.2 respects every party's right to end the relationship, and a client who opts out of AI should still get the full service.

If you already have clients on an older agreement, send an addendum rather than quietly changing practice. A two-line email with the clause pasted in and a "reply yes, or tell me you would rather not" is enough, and keep the reply with your records under Standard 2.4.

What never goes to AI, with the reason for each

A list of prohibitions is easier to follow when each has a reason you can repeat to a client or a supervisor. Here are the five I would put in the agreement as hard limits.

1. Session recordings and transcripts in a general-purpose chatbot. They contain everything, including third parties who never agreed to anything. If you use a note-taking tool at all, the standard to hold it to is the one in AI note takers for executive coaches, where the confidentiality questions are worked through in detail.
2. Anything about health, finances, relationships or legal trouble, even paraphrased. Generalise until it is no longer about anyone, or write it by hand.
3. Messages sent automatically. A tool that texts clients on a schedule without you reading the text is a risk under Standard 2.2, because you cannot know what a model generated at 07:00 on a Tuesday.
4. A client-facing chatbot you have not tested against a crisis. If a client types something alarming, Standard 2.3 expects an agreed route for disclosure. A bot cannot follow it. See whether coaches should build an AI version of their method before you go there.
5. Any account where you have not checked the training and retention settings. That includes a free account you only use "for quick things".

Five checks to run before any message leaves your outbox

Here is the review routine I would build into your week, short enough to do in under a minute per message. It works as a filled-in checklist: the example answers beside each question are for a recap about a worry window.

If any answer is no, fix it or write the message by hand. Keep a log of exceptions: two lines a week. After a quarter, the log tells you which part of your process leaks, which is more useful than any policy.

Choosing the tool: general chatbot or coaching platform

For drafting, a general assistant on a business plan or an individual paid plan with training off is enough. Coach-specific platforms add scheduling, journalling and client portals, and they may host client data themselves. That moves the Standard 2.5 question from "what did I paste?" to "what does this vendor do with everything my clients enter?". Ask the vendor these questions in writing, and keep the answers.

• Where is client data stored, and can I delete a client's data on request?
• Is any of it used to train models, now or in future?
• Who at the vendor can read client entries, and under what circumstances?
• What happens to the data if I close my account?
• If the product coaches clients directly, which parts of the ICF AI Coaching Framework and Standards does it claim to meet, and how do they show it?

The last question is where the framework earns its keep. A vendor that has read it can answer in specifics. A vendor that has not will talk about "bank-level security". For a cost-by-cost view of whether specialist products justify their price over a general chatbot, read are AI tools built for coaches worth paying for.

A week in the life: a practice with fourteen clients

To see how this fits together, take the same illustrative practice with fourteen active clients. On Monday the scheduling tool sends template reminders for the week's sessions, written once with AI and containing no content. Between Tuesday and Thursday the coach holds sessions and, within ten minutes of each, jots three bullets in a private notes app. Friday morning is the batch: with fourteen clients on fortnightly sessions there are seven sets of bullets, each goes into the recap prompt, and the coach reviews seven drafts at about 3 minutes each, adds a personal line, and sends. On Friday afternoon the coach asks AI for 12 generic reflection questions on next week's most common theme, then picks one or two per client and rewrites them.

Total AI-assisted time is about an hour a week (roughly 35 minutes of bullets and review, 20 minutes on the question bank, a few minutes on reminders), and at no point has a client's name, transcript or confidence touched a tool. If you want to see the same discipline applied to the broader note-taking workflow, AI session notes and follow-ups for coaches covers a weekly rhythm for it.

How a leak shows up, and how to catch it

Realistic failures are mundane. Picture a coach who pastes "[first name], who is leaving a job in local government to care for a parent" into a prompt (an illustration, not a real case). The recap comes back repeating the whole sentence in an email that is also copied to an assistant. No hacker is involved. The model faithfully uses what it was given, and the coach does not read closely enough to notice. Nothing in that chain is unusual, which is why the input rule matters more than the output review: the best place to stop a leak is before the paste.

Another quiet failure is drift. After a few months the recaps start sounding alike, the personal line disappears, and clients stop replying. Check your sent folder quarterly. Open ten recaps at random and ask if each could have gone to anyone. If more than two could, tighten the process.

A third is the forgotten account. Coaches test several tools and leave history in each. Once a year, list every AI account you have used for work, delete old conversations, and close the ones you do not need. That is Standard 2.4 in practice: records are disposed of in a way that promotes confidentiality.

Keeping the agreement, your tools and your supervisor aligned

Write a one-page AI use note for yourself: the tools you use, the jobs each does, the settings you verified and on which date, and the sort from the three-column table. Review it every six months or whenever a vendor changes terms. If you have a coaching supervisor, share it. The Code asks you to uphold confidentiality and keep records responsibly, and a supervisor is the person best placed to say when a practice has drifted from that. It also gives you a ready answer if a client, a sponsor or a credentialing reviewer ever asks how you use AI. If you hold an ICF credential, check the current Code text and any credential updates on the ICF site, because this area is moving fast.

If you would rather map this with someone, that is what my AI implementation consultation covers, and you can reach me through the contact page. For the wider picture of AI in a small practice, the AI guide collects the rest of the series.

Questions coaches ask before switching AI on

Do I need client consent before using AI on session material?

Under Standard 1.1 you explain the nature of the work and agree roles, confidentiality and other terms before coaching starts, so consent belongs in the coaching agreement. If you added AI after clients signed, send a short addendum and wait for a yes before any of their material goes near a tool.

Can I paste a session transcript into ChatGPT or Claude?

Treat a transcript as the highest-risk item you hold, because it holds names, relationships and details clients told you in confidence. Most coaches never need to paste one. Write your own three-line recap in plain words, remove identifying details, and give the tool only that.

Is the ICF AI Coaching Framework a rule I must follow?

It was written for providers, developers and organisations building AI coaching, not as a rulebook for a solo coach drafting reminders. The Code of Ethics is the binding document for ICF members. Use the framework as a yardstick when judging an AI coaching product you might recommend.

What if a client tells the AI something alarming between sessions?

Decide this in the agreement before it happens. Standard 2.3 asks for clear agreement about what may need disclosing, such as imminent risk of harm. A tool that sends clients to a chatbot with no human watching cannot honour that, so keep client-facing AI to prompts you wrote and reviewed.

Further reads

• How Coaches Use AI to Build Workbooks, Exercises and Programmes — Build the exercises your reminders point to.
• Life Coach Questionnaire for Clients: Build and Analyse It With AI — Intake questionnaires and summarising answers.
• How to Mention AI Use in Client Contracts and Proposals — Wording for telling clients you use AI.
• Is It Safe to Put Customer Data Into ChatGPT? — The data-handling basics behind every rule here.
• How Coaches Use AI to Qualify Discovery Call Bookings — Use AI before the first session, not just between.
• AI 360 Feedback for Coaches: Collect, Theme and Report — Another coaching job where confidentiality shapes the tool.
• AI for Fitness Trainers: Client Check-Ins Without Medical Advice — A check-in script, a never-answer list, fixed red-flag hand-off messages and a 12-message test for trainers who use AI between sessions.
• Life Coaching Rates During Automation: How to Reprice Packages — How to reprice life coaching packages once AI handles prep and admin: hours-based versus milestone-based structures, with a worked example.
• Best AI Model for Life Coaching: What Session Notes and Privacy Need — How a life coach should choose between ChatGPT, Claude and Gemini for session prep and follow-ups, with privacy settings, ICF Code limits and a test routine.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ICF Code of Ethics (in effect from 1 April 2025) and the ICF Artificial Intelligence (AI) Coaching Framework and Standards V1.01 (coachingfederation.org); ChatGPT Business and Claude Team pricing pages and OpenAI's data controls FAQ, checked November 2026.