In November 2024 the International Coaching Federation published its AI Coaching Framework and Standards. It is a careful, useful document. It is also 38 pages long and written mostly for the people who build AI coaching systems, not for the coaches who use them.
I build AI coaches for coaches, so I read it closely. This is a plain English guide to what it says, who it applies to, and what it means if you are thinking about an AI coach for your own clients. It is my reading, not an official ICF summary, so for anything that matters to your practice, go to the original.
The short answer
The ICF AI Coaching Framework and Standards (version 1.01, published November 2024) sets out what a responsible AI coaching system should do. It groups 13 sets of standards into six areas: ethics and coaching mindset, the coaching relationship, communication, learning and growth, testing, and technical factors such as privacy and security. Most requirements have a basic level, and many also have an advanced one.
For a coach, the practical points are these. Clients should know they are talking to AI before it starts. The system should state its limits, say it is not for mental health and point to help, and give clients a way to reach a human coach. Clients should opt in and consent to how their data is used, and someone should be able to say who can see that data and where it is kept.
The standards are guidance, not law. Providers can check themselves against them or work with a third party, and they can say which parts they meet rather than all or nothing.
Who the ICF AI coaching framework applies to
The framework sorts AI tools in coaching into four types:
- Scheduling tools, which help with bookings, reminders and follow ups.
- Data processing tools, which analyse things like feedback or session data for the coach to use.
- Interactive tools, which work with clients directly, for example goal prompts and habit nudges.
- Conversational tools, which hold coaching conversations with clients, such as chatbots, voice assistants and avatars.
The first two support the coach. The last two coach the client, and that is where the standards focus, because they handle more sensitive conversations and can reach far more people than any human coach.
So if you use AI to tidy your session notes, the standards are mostly not about you. If you are putting an AI coach in front of your clients, they are. Strictly, it is the system that meets the standard or not, not the coach. But your clients will see your name on it, so in practice the questions land with you.
The six areas in plain English
1. Foundation: ethics and a coaching mindset
This is the heart of it. The system should tell clients it is not human before it starts coaching them, and at the advanced level, remind them at least once in every session. It should state what it cannot do. The provider should explain in general terms how it works and how it makes decisions, without giving away trade secrets. Clients should be told about possible bias. There should be a clear record of who can access client data and where it is stored. And clients should agree to AI coaching before it begins.
On mindset, the system should be mostly non-directive. It partners with the client rather than telling them what to do, which is the same standard you hold yourself to.
2. Co-creating the relationship
Before coaching starts, the system should tell clients what it offers and how it might help, and ask them to opt in. It should ask before moving into a new kind of help. Clients should be able to find out who made it and what it is based on, and whether it is built for a particular area of coaching.
Two requirements here matter more than any others. The system should tell clients how to reach a human coach, and at the advanced level, let them book or contact one. And it should say clearly that it is not meant for mental health support, and point clients to help. At the advanced level, it should watch for signs of risk in what a client says and, if it sees them, make clear what it cannot help with.
3. Communicating effectively
The system should use a neutral, professional tone, let clients review and change what they have said, and adapt to what it hears rather than following a fixed script. It should ask open questions, offer other perspectives and explain things when asked.
4. Cultivating learning and growth
This covers what good coaching does over time: helping clients set and track goals, explore alternatives, check that goals are realistic, clarify the real problem, think through options and consequences, and reflect on their own progress. The system should give regular feedback along the way.
5. Assurance and testing
The provider should collect feedback on whether it actually works, and back up any claims it makes. It should monitor accuracy and bias, test with a varied group of people, and keep checking the material and rules it was trained on. At least one coaching expert should review the design. It should be easy to use, written in clear, simple language, and free of discriminatory content.
6. Technical factors
The basics of any system that holds sensitive conversations: encrypted data, a secure sign-in, being available when clients need it, clear information on how data is managed, and consent from the client for their data to be processed.
What it means if you are adding an AI coach to your practice
Read as a coach rather than a developer, the standards turn into a short list of questions. Ask them of any provider, including us:
- Does it tell clients it is AI before the first conversation, and keep that visible?
- Does it say what it cannot do, in words a client will actually read?
- What happens when a client mentions something serious? Does it say it is not for mental health, point to help, and tell you?
- Can clients reach you, the human coach, from inside it?
- Do clients opt in, and agree to how their data is used?
- Who can see the conversations, and where are they stored?
- Does it coach or does it lecture? Ask it a few real client questions and see whether it asks as well as tells.
- Who checks its answers over time?
One requirement is easy to miss: the design should be reviewed by at least one coaching expert. For an AI coach trained on your material, that expert is you. Before any client sees it, ask it the questions your clients ask, and the awkward ones, and change what it gets wrong.
If you hold an ICF credential, nothing about your own ethical obligations changes because a tool is doing part of the work. Confidentiality and clear agreements with your clients still sit with you. If an organisation pays for the coaching, AI for executive coaches has a confidentiality checklist for that three way agreement.
How this maps onto our AI coaches
Several of the standards match how we already build. Our AI coaches answer only from the coach's own material, passes anything personal or worrying back to the coach with context, and is set up together with the coach, who reviews its answers before launch. The conversations belong to the coach, who can export them and leave at any time.
The framework does not describe an ICF approval for individual products, so be careful with anyone who implies one. Ask the questions above instead, and judge the answers.
Other ICF resources worth knowing
The ICF has published two practical companions since the standards came out:
- A practical guide to integrating AI and coaching (January 2025), for coaches using AI and for the people building it.
- A self-scoring worksheet (February 2025), for checking your own practice against the standards.
Questions coaches ask
Are the ICF AI coaching standards mandatory? No. They are guidance. Providers can check themselves against them or work with a third party, and can say which parts they meet.
Do they apply if I only use AI for notes and admin? Mostly not. The standards focus on AI that coaches clients directly. Your usual care with client data still applies.
Can an AI coach be ICF accredited? The framework talks about providers checking themselves or using a third party. It does not describe the ICF approving individual products.
Does the ICF say AI will replace coaches? No. The framework sees AI taking on routine work so coaches can spend more time on the complex, human side of coaching, while also recognising that some coaching services may be run entirely by AI. I have looked at what the research says in will AI replace coaches?
Where does an AI coach trained on my material fit? It is a conversational tool in the framework's terms, so the standards apply to it. That is a good thing: the questions above are exactly the ones your clients would want you to have asked. There is more on how these coaches work in AI clone for coaches: what it is and what it costs.
Where to start
Download the standards, read the six areas once, then put the eight questions above to whatever AI tool your clients will touch. If the answers are clear and short, you are in good hands.
If you'd like to see what this looks like for your own practice, I'll build a companion from your website, free. See your own companion
Robert, founder of Thinkable Systems