Do You Have to Tell Clients You Use AI? Standard 2.5 in Practice

If a quiet worry has surfaced - that you adopted an AI note-taker or a prep assistant somewhere along the way and never mentioned it to a client - this article is the place to put that worry down and replace it with a defined obligation. The disclosure question routes through the broader picture in the full guide to AI in your practice, but the practical mechanics live here: what counts, what to write, when to ask again, and what happens if a client says no.

The ICF Code of Ethics, Standard 2.5, is one sentence long. If you have been carrying a quiet worry that you should have told a client about the AI tool in your practice, that one sentence is the whole of what you are accountable to under the Code - and it asks for something more precise, and less punishing, than the worry suggests. Read it before you decide you have done anything wrong. In the 2025 ICF Code of Ethics, 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.”

That is the standard. Notice what it does not say. It does not ask you to have disclosed every tool retroactively, to be a data-protection expert, or to pass an audit. It asks for awareness, for clear and appropriate boundaries, and - read alongside the disclosure expectations the rest of the Code carries - for honesty with the client about the interactions a tool introduces. The verb is about awareness and disclosure, not certification.

The obligation is also fixable. The dread that brought you here is unbounded - “have I been unethical?” The standard is bounded. A coach who has not yet disclosed an AI tool has a forward-looking task, not a permanent mark. Standard 2.5 does not punish the gap. It asks you to close it.

One thing this article is, and one thing it is not. It operationalizes an ICF ethics standard for practicing coaches - it is professional education. It is not legal advice. Standard 2.5 is an ethics standard, not a data-protection statute, and a coach with EU clients, organizational contracts, or a complex jurisdiction has obligations beyond it. Where the line touches data law, check with counsel. What follows is what the standard asks of you in practice, in the order you need it.

What Standard 2.5 Actually Asks of You

Standard 2.5 places the obligation on transparency, which is the part of this you fully control. It does not prohibit AI tools, and it does not hand you a permission slip. It asks you to be aware of the boundaries a tool changes, and to set them clearly with the client. The interaction a recording tool introduces, or a transcription service, or an assistant that processes the session - that is an interaction “physical or otherwise” with your client, and the standard asks you to govern it openly.

The phrase doing the most work is “clear, appropriate, and culturally sensitive boundaries.” That is a judgment phrase, and it is one by design. ICF writes principles a credentialed coach applies, not a checklist a coach completes. The standard does not specify a format for disclosure, a moment for it, or a wording. It does not tell you whether the disclosure belongs in the agreement or in conversation, or both. Reading it honestly means admitting it is not more specific than it is.

A seam runs through this article, and it is worth naming. The quoted sentence above is ICF's. Everything that follows - the test for what counts, the agreement language, the re-consent triggers - is Tandem's operationalization of the standard for working coaches. It is a defensible reading, but it is a reading. The standard commits you to awareness and clear boundaries; the practice that satisfies that commitment is what this article supplies.

A standard written as a principle is not a vague standard. It is a standard that trusts you to be a professional. The discomfort of not having a script handed to you is the discomfort of being asked to use judgment - which is the whole of what coaching is. Standard 2.5 sits inside a wider context, too: ICF's AI Coaching Framework, published in November 2024, names transparency and disclosure as one of its six domains. A coach who wants the full framework picture will find Standard 2.5 in the full ICF AI Coaching Framework context there. For the disclosure question itself, the standard above is enough to work from.

What Counts as Disclosable - and What Doesn't

Here is the question most coaches actually arrive with: surely a note-taker, or a prep tool, or a transcript summarizer is too minor to mention? The answer is a test, not a list - and the test does not sort tools by how significant they feel. It sorts them by what they touch. Two questions, asked in order.

Question 1, the early disqualifier: does the tool touch identifiable client data, or the coaching conversation itself? If a tool records, transcribes, or processes the session, or stores anything that identifies the client, it is disclosable - that is the end of the evaluation, and the answer is yes. If a tool never touches client-identifiable data, it is not a disclosure event. The test sorts on data and on the conversation. It does not sort on whether the tool felt important enough to mention.

This is the point where the technical sorting matters, so let it be mechanical for a moment. A session note-taker that joins the call touches the conversation directly - disclosable. A transcript summarizer processes identifiable session data - disclosable. A scheduling assistant that holds the client's name and email touches identifiable data - disclosable, though, as the next question shows, the disclosure can live in a privacy policy rather than in a session conversation. A coach using a general AI tool to think through their own stuck reaction to a client, entering no name and no identifying session detail - that does not touch identifiable client data, and it is not a disclosure event at all. Once Question 1 is the lens, the sort stops being a feeling and becomes a yes or no.

That last case is worth naming as a clean, confident use. A coach using AI for their own development - working through a framework, examining their own pattern, with no client detail entered - is on solid ground and owes no one a disclosure. The test is not “disclose everything AI-shaped.” It is “disclose what the standard is actually about.”

Question 2: if a tool is disclosable, what level of disclosure does it need? Not every disclosable tool needs the same treatment, and treating them all the same is its own failure. A back-office scheduling tool that holds a client's contact details is disclosed adequately in a written privacy policy - the client can read it, and that is proportionate to what the tool does. A transcript summarizer that processes session content belongs in the coaching agreement as a named clause. A note-taker that joins the live session needs more than a written line - it needs a spoken disclosure and the client's consent before it runs. Question 2 maps each disclosable tool to the disclosure level it actually requires.

Knowing whether a tool is disclosable in the first place starts upstream of this test. A tool's data behaviour is not always visible from the outside - which servers hold the recording, whether the vendor trains on transcripts, how long data survives. Knowing whether a tool is disclosable starts with the vetting test, because the vetting test is what surfaces whether a tool touches identifiable data - which is Question 1. Run the vetting test first; the disclosable test reads its results.

The hardest case will not sort cleanly, and an honest test says so. Consider the reflective use again - a coach processing a stuck client between sessions with an AI partner. If no identifying detail is entered, Question 1 says not disclosable. If the coach pastes in session detail, names, or specifics, the same use becomes disclosable, because now identifiable data is in the tool. The test does not resolve the coach's discretion for them. It tells them exactly which choice makes the difference: what you type in decides which side of the line you are on.

Consider two coaches looking at the same set of AI tools. The first treats everything as disclosable and reads the client a long list at intake - the scheduling software, the email client's smart-compose, the note-taker - until the client's attention drifts and the genuinely important disclosure, the note-taker that joins the live session, is buried in the middle of a recital. The second runs the two questions, discloses the note-taker clearly and in a spoken moment, and puts the scheduling tool in the written privacy policy where it belongs. Neither coach is hiding anything. One has confused volume with transparency. Precise disclosure - the note-taker named clearly, the scheduling tool placed proportionately - is what the standard is actually after.

The Words to Put in Your Coaching Agreement

This is where the obligation becomes a paragraph you can write. Standard 2.5 asks for clear boundaries; the coaching agreement is where the written record of those boundaries lives. A disclosure clause does not need to be long, and a disclosure the client cannot understand is not really a disclosure.

A workable disclosure clause covers, in plain language, five things: that AI tools may be used in the course of the engagement; what kind of tools - note-taking, transcription, scheduling; what data those tools touch; where that data is processed and how long it is kept; and how the client can ask questions or decline. An illustrative version, kept deliberately short, reads something like: “In the course of our work I may use AI-supported tools for tasks such as scheduling and session note-taking. Where a tool records or processes our sessions, I will tell you specifically which tool, what it does with the recording, and how long that data is kept. You can ask me about any tool at any time, and you can decline the use of a recording or transcription tool without affecting our work together.” That is the shape. It is illustrative, not a finished clause.

One caution on templates. The agreement language is a starting point you adapt to your actual tool stack and your own voice. A clause pasted unread is its own kind of non-disclosure - if a client asks you a question about a sentence in your agreement and you cannot answer it, you did not really disclose anything; you copied something.

A disclosure clause pasted unread is not a disclosure. It is a sentence the client cannot question and the coach cannot defend.

The written clause is also not the whole of disclosure for in-session tools. The agreement is the record. For a tool that joins the live session - a note-taker, a recorder - the consent is a spoken moment at intake: a short, plain exchange where you name the tool, say what it does, and confirm the client is comfortable before it runs. The clause documents; the conversation consents. Both belong in the practice.

The standing caveat applies here too. The disclosure language is a practitioner starting point, not a complete legal instrument. A coach with organizational clients or clients in a regulated jurisdiction should have counsel review the agreement - the clause handles the ethics obligation under Standard 2.5; it does not, on its own, handle data law.

When You Need to Have the Conversation Again

Disclosure is not a one-time signature, and that fact tends to produce its own anxiety - the sense that you must re-disclose constantly, check in before every session, never let the client forget there is software involved. That is not what the standard asks either. Re-consent is triggered by a small, nameable set of changes. Outside those triggers, the agreed disclosure stands.

The first trigger is a new tool. A client consented to the tools you disclosed at intake, not to tools you adopt three months later. When a new tool enters your practice and touches the client's data or the conversation, the client mid-engagement is owed a fresh disclosure before it runs in their session.

The second trigger is a material change in an existing tool's data handling. A vendor changes its retention period, begins training its models on customer data, or moves data residency to a different country. The tool the client agreed to is, in a meaningful sense, no longer the same tool. A coach who learns of such a change re-discloses it.

The third trigger is a change in how the tool is used. A note-taker the client agreed to for the coach's private notes is now generating a summary shared with a third party - a sponsor, an organizational buyer. The tool did not change; the use did, and the use is what the client consented to.

The fourth trigger is the returning client on an old agreement. A client comes back for a new engagement on an agreement signed before you adopted any AI tools. That agreement is simply silent on AI. Before the next AI-touched session, a fresh disclosure is owed - the old signature cannot cover tools that did not exist when it was given. Session recording tools are their own re-consent category here, because they are the most common mid-engagement adoption and the most data-heavy: session recording tools trigger specific disclosure obligations that the note-taker article details in full.

Picture how a trigger plays out. A coach is mid-engagement with an established client and adopts a new transcript-summarizing tool. The original agreement disclosed a note-taker but not this new tool, which sends data to a different vendor with a different retention policy. The coach recognizes the first trigger - a new tool that touches the conversation - and raises it at the start of the next session in two plain sentences: here is a new tool I have started using, here is what it does with our session, are you comfortable with that. The client agrees. The conversation took ninety seconds. The trigger was real and the obligation finite - a named change and a short exchange, not a renegotiation.

What Happens When a Client Says No

The largest fear underneath the disclosure question is rarely stated plainly: that disclosure invites refusal, and refusal costs the engagement. So name it plainly. A client declining an AI tool is a coaching outcome, not a coaching failure. The coach adapts the tooling, and the relationship continues unbroken.

The practical adaptation is usually small. If a client declines a note-taker, the coach takes manual notes - the way coaches did for decades. If a client declines a transcription tool, the coach turns it off, or uses a tool that does not touch the session. The engagement does not depend on the AI tool. It depends on the coaching, and the coaching is unaffected by which note-taking method is in use.

There is a deeper point here, and it inverts the fear. A client who was asked, and who could have said no, has learned something about the coach: that this coach takes their confidentiality seriously enough to ask. The disclosure conversation builds trust whether the answer is yes or no. The client who says yes does so knowingly. The client who says no has just watched the coach honour a boundary. Both outcomes strengthen the relationship. The version that damages trust is the one where the client was never asked and discovers the tool on their own.

The client-says-no path is also a quiet check on the coach. If an engagement genuinely cannot proceed without one specific AI tool, that dependency is worth examining - it suggests the tool has become load-bearing in a way a coaching engagement should not be.

Asking permission is not a risk to the relationship. It is part of the relationship. A client who says no has not rejected you - they have trusted you with a boundary, and holding that boundary well is the coaching.

Here is the task, and it is smaller than the worry that brought you here. Open your coaching agreement template tonight. If it says nothing about AI tools, add the disclosure paragraph - the template gives you the words to start from. Then make a short list: every AI tool currently in your practice, sorted by the one question - does it touch the client's data or the conversation? The tools that do are a conversation you owe your next client at intake. That is the whole of it. A tool you have not disclosed is a conversation you have not had yet - and now you know exactly how to have it.

This article quotes the ICF Code of Ethics, Standard 2.5, and references the 2025 ICF Code of Ethics and the ICF AI Coaching Framework (November 2024). It is professional education, not legal advice. Standard 2.5 is an ethics standard, not a data-protection statute - a coach with organizational clients or clients in a regulated jurisdiction should have a coaching agreement reviewed by counsel.