What ICF Actually Says Coaches Can and Can’t Do With AI

In November 2024, ICF published an Artificial Intelligence Coaching Framework and Standards. It is built as six domains, and most credentialed coaches have not read it. The ones who have often open it looking for a verdict - a yes or a no about whether a particular tool is allowed. That is not what the document is. It does not hand a coach a ruling. It hands six domains of considerations, and it sits inside the practitioner’s guide to where AI belongs as the document the rest of those decisions reference. This article reads all six domains, in ICF’s own words, and shows what each one does and does not ask of you.

Before the six domains, it is worth being precise about what kind of document this is, because the misreadings start here. The framework was published in November 2024. It sits alongside the ICF Code of Ethics and the ICF Core Competencies - it does not replace either. And it is, in its own words, a standard for an AI coaching system. The framework states plainly: “This Standard relates to an AI Coaching System. It is the System that conforms or does not conform with the Standard.”

That sentence resets the whole reading. The framework is addressed to the people who build, buy, and use AI coaching products. It is not a rulebook telling a practicing coach which tools they may bring into their own practice. A working coach reads it the way they would read a quality standard for an instrument they rely on - to understand what ICF expects of AI in coaching, then to apply those principles by analogy. ICF wrote it as principles for a professional to interpret, not procedures to follow. That is the same design as the Core Competencies, and it is deliberate.

So there are two misreadings to set aside. The first reads the framework as a prohibition list and is relieved or alarmed to find none. The second reads it as blanket permission and is sharpened, on closer reading, by how many real questions it raises. Neither is the document. The framework names a structure for thinking, and the rest of this article walks all six domains in ICF’s own words.

A. Foundation - What ICF Says, and What It Asks of You

The first domain is Foundation, and the framework describes its purpose directly. In ICF’s words:

“The foundation elements relate to ethics and the coaching mindset. These two aspects are crucial to ensuring that the right values are embedded in all design decisions and core functionality of the system.”

Foundation holds two sub-standards: A.1 AI Ethics and A.2 Embodies a Coaching Mindset. On ethics, the framework is specific about what it asks of an AI coaching system. It states that “AI coaching applications should be designed in alignment with codes of conduct that ensure an automated system is able to respond to situations in an ethical way that supports and complements the ICF Code of Ethics.” Read that carefully. The framework does not replace the Code of Ethics. It asks AI systems to be built so they support and complement it - the Code remains the reference point.

The second sub-standard, Embodies a Coaching Mindset, names what the system should resemble: “A coaching mindset is one that is open, curious, flexible, and client-centered. These are human attributes, and the AI coaching platform should mimic these characteristics.” The verb there is worth pausing on. The framework says an AI platform should mimic human attributes - it is careful not to claim the platform has them. That is the document being precise, and it matters.

For a practicing coach, the Foundation domain does work even though it addresses systems. It tells you what ICF expects of any AI you let near coaching: transparency, active work to minimize bias, and a design that complements rather than competes with the Code you are credentialed under. When you evaluate a tool, Foundation is the question of whether its design respects the ethical frame your credential commits you to. The framework asks that of the system; you apply the same question to your choice of system.

B. Co-Creating the Relationship

The second domain concerns the relationship itself, and the framework opens it with a candid admission about where AI struggles:

“Research has established the importance of a coaching relationship. Many elements combine together to create the depth of relationship necessary for a client to feel safe in the coaching relationship. While the development of a coaching agreement may be relatively simple with AI, a depth of presence and creation of a trusting and safe relationship may be more difficult.”

That is the framework naming its own hard problem. It says the easy part - agreements - is relatively simple for AI, and the hard part - presence, trust, safety - “may be more difficult.” The domain holds three sub-standards: B.3 Establishes and Maintains Agreements, B.4 Cultivates Trust and Safety, and B.5 Maintains Presence.

The Maintains Presence sub-standard connects most directly to a working coach’s daily judgment. The framework asks that an AI system “should not be overly anthropomorphized or misleading in presentation to avoid unsettling the client.” It is telling system designers not to dress an algorithm up as a person. For a human coach, that principle operationalizes a line you already hold: the relationship belongs to the human, and a tool that blurs that belonging is a problem.

What the domain does not do is hand a coach a verdict on any particular product. It describes what trust, safety, and presence require of an AI system, and admits openly that those are the hard ones. The reading you take from it is not “avoid tool X” but “here is what ICF says the relationship depends on - hold your tools against it.”

C. Communicating Effectively

The third domain covers how an AI coaching system communicates, and the framework frames it this way:

“Effective communication in an AI coaching system involves emulating the app doing active listening to fully understand the client and then use that to evoke client awareness through diverse methods such as personalized assessments, reflective questioning, and scenario simulations, coupled with detailed feedback and progress tracking.”

Notice the word the framework chooses: emulating. It does not say the app listens. It says the app emulates active listening. That is the document staying honest about what the technology is doing. The domain holds two sub-standards: C.6 Listens Actively and C.7 Evokes Awareness - the same competency names a credentialed coach already knows from the human Core Competencies.

On listening, the framework says active listening “involves understanding both the spoken and unspoken aspects of client communication, contextualizing their words for meaningful interactions.” On evoking awareness, it says an AI application “can enhance client self-awareness and insight using various techniques like powerful questioning, metaphors, and analogies” and “should offer a range of context-specific stimuli, such as open questions and moments of silence.” A coach reading those lines will recognize their own competency model reflected back, applied to software.

This is the domain where the framework’s relationship to the Core Competencies is clearest, and where a practicing coach should be most careful. The framework describes what a system should emulate. It is not granting any coach permission to let a tool perform their listening or their questioning. The competency, for the human coach, remains the human coach’s. Reading the verb - emulate - keeps that distinction intact.

D. Cultivating Learning and Growth

The fourth domain is the one the framework calls the core of the work. In ICF’s words:

“Cultivating learning and growth is the core of coaching and can be broken into two components; facilitating client growth, and reinforcing client growth.”

The domain holds two sub-standards: D.8 Facilitates Client Growth and D.9 Reinforces Client Growth. The framework describes facilitation as helping clients “convert insights into actionable goals while fostering client independence” - including setting and adjusting goals, “facilitating storytelling and meaning-making,” and “supporting clients in re-evaluating their goals as circumstances change.” Reinforcement covers “tracking and validating progress,” “timely reminders,” and helping clients “celebrate their successes to sustain growth.”

Two phrases in that domain deserve a coach’s attention. The first is “fostering client independence” - the framework names, as a property the system should have, that the client should grow less dependent over time, not more. The second is “meaning-making,” which the framework lists as something the system should facilitate. A credentialed coach knows how much weight that word carries, and how much of meaning-making is relational. The framework asks AI systems to support it; it does not claim they complete it.

Domain D is also where the framework completes its mirror of the human Core Competency structure. ICF states it directly: “The first four Domains reflect the ICF Core Competency Framework as used for human coaches. The same principles have been applied here, although resulting in different criteria.” That sentence is the key to reading the whole document. Domains A through D are the competency model you trained against, re-expressed for AI systems. A coach who knows the competencies already holds most of the framework’s first four domains - they are reading a familiar map in a new application.

E. Assurance and Testing

The fifth domain is one of the two the framework describes as new. Where Domains A through D mirror the human competencies, ICF says of Domain E:

“Quality assurance and testing are crucial to ensure that the system is effective and achieves its intended objectives. The earlier sections A-D focuses on the kinds of content, capabilities, and behaviors of the systems. This section instead looks at what can be done to validate that the AI system performs effectively.”

The domain holds two sub-standards: E.10 Coaching Reliability Measures and E.11 System Usability. Reliability, the framework says, means an AI coaching application should “quantitatively demonstrate system reliability” - collecting “evidence of effectiveness, which can range from client feedback to formal experiments, and potentially comparing it to the efficacy of human coaches.” Usability covers whether a system is “user-friendly and not overly complex, with content that is easily understandable.”

This domain has no equivalent in the human Core Competencies, and that absence is informative. ICF is not asking a human coach to run formal experiments to prove they coach effectively. It asks that of AI coaching systems, because a system makes a claim - it works - that should be tested rather than trusted. For a practicing coach, Domain E licenses a healthy skepticism. When a vendor says their tool improves outcomes, Domain E is ICF telling you it is reasonable to ask: where is the evidence, what was measured, against what comparison. The framework does not resolve any vendor’s claim; it tells you the question is legitimate to ask. That posture travels with extra weight into reflective or developmental AI use, and it is one strand of what the supervision domain in the AI Coaching Framework takes up in depth, alongside the human oversight no automated reliability measure replaces.

F. Technical Factors

The sixth and final domain is the other new one, and the framework introduces it plainly:

“While not specific to coaching per se, certain requirements are necessary for any consumer product storing personal and potentially sensitive information. AI Coaching Systems should meet minimum security and privacy requirements.”

Domain F holds two sub-standards: F.12 Security and Privacy and F.13 Resilience and Accessibility. On security and privacy, the framework grounds the standard in “the CIA triad: confidentiality (preventing unauthorized data access), integrity (protecting data from unauthorized alteration), and availability (ensuring system access for authorized users).” It asks that data privacy practices adhere “to varying data protection laws across jurisdictions” and that systems “minimize personal data.” On resilience and accessibility, the framework asks systems to “incorporate adaptive technologies and design principles that cater to a wide range of disabilities.”

Domain F is the most concrete in the whole framework - it is where coaching meets data law. A coach evaluating a note-taker, a transcription service, or a scheduling assistant is making a decision Domain F governs: where client data lives, who can access it, how long it survives, whether the vendor trains on it. The framework does not vet any product for you. It names the properties a responsible system should have, so you know what to look for.

That completes the framework walk. You have now seen all six domains as ICF wrote them - Foundation, Co-Creating the Relationship, Communicating Effectively, Cultivating Learning and Growth, Assurance and Testing, and Technical Factors. The six together are the map. They are not six separate rules to pass or fail; they are six areas of consideration a credentialed coach is expected to think across whenever AI enters the picture.

Standard 2.5: The One Standard That Names AI Directly

The AI Coaching Framework is a standard for systems. The document that binds a human coach is the ICF Code of Ethics, and there is one standard in the 2025 Code that the disclosure obligation routes through. Standard 2.5 commits the coach to:

“Be aware of and set clear, appropriate, and culturally sensitive boundaries that govern interactions, physical or otherwise, that I may have with my clients or sponsors.”

Standard 2.5 does not name AI. It does not have to. An AI tool placed inside a coaching relationship - a recorder, a transcription service, an assistant that processes the session - introduces an interaction “physical or otherwise” with the client, and the standard asks you to be aware of it and to set clear boundaries around it. The phrase carrying the weight is “clear, appropriate, and culturally sensitive boundaries.” That is a judgment phrase, written for a credentialed coach to apply, not a checklist to complete.

The relationship between the framework and the Code is worth stating cleanly. The framework helps a coach think - it is guidance, addressed to systems, useful to a coach by analogy. The Code of Ethics is the standard a human coach is actually held to. The two are not in tension. The framework operationalizes, for AI systems, the same disposition Standard 2.5 asks of the coach: be transparent, set clear boundaries, govern the interaction openly. For the practical mechanics of how a coach satisfies that obligation - the agreement language, the intake conversation, the re-consent triggers - Standard 2.5 in practice is the working guide.

One note, kept brief and non-interrupting: this article reads ICF documents for practicing coaches as professional education. It is not legal advice, and Standard 2.5 is an ethics standard, not a data-protection statute.

How a Coach Actually Uses This

Six domains and one standard. The reasonable question a working coach is left with is what to actually do with all of it on a Tuesday, when a real decision is in front of you and there is a client at four o’clock.

Start by being honest about what the framework is not. It is not a list of approved tools, and a coach should not wait for one. It gives you the vocabulary and structure ICF uses, and leaves the decision about your own practice to your judgment - which is exactly what a credential is supposed to mean. The value of the six domains is not that they produce an answer. It is that they make your thinking complete: a sequence you walk so that no domain gets skipped.

Picture a coach who has decided to use an AI tool to prepare reflection prompts before sessions, and wants to know if they are “allowed.” The framework does not answer that question, and it should not. What it does is give the coach a walk. Foundation: does the tool’s design respect the ethical frame, and is it transparent about what it is. Co-Creating the Relationship: does anything here touch the client’s trust or the presence I bring into the room. Communicating Effectively: am I selecting and adapting the prompts with judgment, or about to deliver the tool’s output as my own. Cultivating Learning and Growth: does this serve the client’s growth and independence. Assurance and Testing: what does the tool claim, and is that claim evidenced. Technical Factors: where does any client data go, and is it secure. The coach is not looking up a rule. They are working a map.

ICF did not publish a list of approved AI tools, and that is the feature, not the gap. A list ages the moment a vendor changes its terms. Six domains are the questions a credentialed coach was always going to have to answer.

That walk is also where the framework hands off to the rest of a coach’s decision-making. AI-assisted preparation of reflection prompts, reviewed and adapted before a session, is a use the framework’s domains clearly support - the coach’s judgment stays in the room, the tool supplies raw material outside it. Other uses raise harder questions. When a coach is weighing the line between a tool serving a competency and a tool standing in for it, the competency line operationalized takes Domains B and C down to the level of individual tools. And the framework applies with particular weight when a session is submitted as evidence for a credential - the credentialing implications of the framework are their own decision worth understanding before any session is recorded for an assessor.

The framework, used well, is a decision instrument - not a document to file. It does not end the thinking. It tells you which thinking cannot be skipped.

This article quotes the ICF Artificial Intelligence Coaching Framework and Standards (V1.0, November 2024) and the 2025 ICF Code of Ethics, Standard 2.5. It is professional education, not legal advice. The framework is a standard for AI coaching applications, addressed to developers, buyers, and users of AI coaching products; a human coach applies its principles by analogy. The framework is dated and will be revised - a coach citing it should confirm they are citing the current version. Where ICF’s documents are silent on a question, this article says so rather than supplying a ruling ICF did not write.