Any business owner whose firm uses automation with clients eventually faces one question: do we tell them? It comes up in agencies, clinics, advisory firms and coaching practices alike, and it lands hardest wherever the relationship is the product.
The instinctive answer is yes, obviously, transparency. The research says that instinct will cost you, which is genuinely awkward, and I want to work through it honestly rather than land on the comfortable conclusion.
The finding nobody wants
Schilke and Reimann ran thirteen preregistered experiments on voluntary AI disclosure. Trust went down in every single one.
The study closest to the daily reality of a service business involved a scheduling email from a colleague. When she mentioned that she used an AI tool to help manage her calendar and send scheduling emails, participants trusted her less than when the identical email carried no disclosure. Not dramatically less. Measurably and consistently less.
The mechanism the authors identify is not competence. Nobody thought she was worse at her job. It was legitimacy: something about announcing the tool made the interaction feel less properly hers.
Their own framing of it is worth sitting with. The people who try to be trustworthy by disclosing end up trusted less for having done it.
The finding that saves you
The same body of work contains the counterweight, and it is decisive.
Being exposed by somebody else is worse than disclosing. Substantially worse. So the choice is not between disclosure and safety. It is between a small, controlled cost you pay on your own terms and a large one you pay on someone else's.
Add the Pew survey to this and the picture closes. In June 2025, 5,023 US adults were asked how they would react on discovering AI was involved in various things. Every scenario leaned negative, including a speech they had already said they liked, which 71% said would lower their view of the candidate. The judgment attaches to the discovery, not to the quality. And people are better at spotting it than most owners assume, a question I dig into in can people tell if your writing is AI.
So the actual question is not whether to disclose. It is how to disclose in a way that does not read as a confession.
Why most disclosure reads badly
Because it is written as an admission, appended to something personal, at the moment of contact.
"This message was drafted with AI assistance" at the bottom of a warm note to a client does maximum damage for minimum benefit. It attaches the machine to precisely the artifact whose value was that a person meant it, and it does so at the moment they are reading it.
The version that works is a standing policy, published where somebody can find it, written in advance rather than attached to a specific message. That does three things. It moves the disclosure away from the intimate artifact. It reads as a considered position rather than a caveat. And it means nobody can ever discover something you did not say.
The policy I would publish
Written for a client to read, not for a lawyer. If you have a team, it covers everyone who speaks to clients, not just you.
What never happens. The conversation where your reading of a situation is the work is a person, always. Nothing writes the substance of what you think about a client's situation. If a client asks a question, the answer comes from someone who knows them.
What might. Scheduling, reminders, notes and their retrieval, the structure of routine messages, keeping track of what was promised and when. The infrastructure of remembering, essentially. I have laid out my own version of this line in where I let AI near the business, and where I do not.
Whose words. Anything that goes out in your voice is built from your material, your language and your way of explaining things, and you can decide per message type whether you approve it before it leaves or let it run. That is much easier if you have a written tone of voice document to build from.
What a client can always ask. Whether a given message came from a person, and get a straight answer. That single sentence does more work than any amount of general reassurance, because it converts an abstract worry into an available action.
Publish it once, on a page, in plain language. Then behave in a way that would not embarrass you if a client read it aloud.
What the professional bodies actually say
Worth clearing up, because it is widely misreported. The two bodies below matter most to coaches and clinicians; if you are in another profession, the principle carries even where the wording does not.
The ICF Code of Ethics, Standard 2.5, requires a coach to fulfill their ethical and legal obligations to clients "directly and through any technology systems I may utilize", naming artificial intelligence among those systems. It does not contain a general duty to disclose AI use, despite a great many blog posts saying it does. What it refuses to allow is technology as a place where responsibility disappears.
The ICF's separate AI Coaching Framework and Standards is strict where it should be. Element A.1.1 requires that an AI coaching system "shall indicate to the Client that it is not human prior to the provision of Coaching Services". That is about systems that deliver coaching, not about a human coach whose reminders are automated. The line the standard draws is the same one above.
The American Psychological Association's November 2025 health advisory tells people not to use chatbots or wellness apps as a substitute for care from a qualified professional, and tells clinicians to ask patients about their AI use rather than wait to be told.
If you are in a regulated profession, your own body's current guidance outranks everything in this post, and it is worth reading directly rather than through anybody's summary.
Questions owners ask
Do I have to disclose AI use to my clients?
Legally it depends on your jurisdiction and your profession, and in many cases there is no general requirement. Ethically, the practical standard is whether being found out would embarrass you. If it would, disclose. If it would not, a published standing policy is enough and is better received than a per-message note.
Where should the policy live?
On its own page, linked from your site footer and mentioned once during onboarding. Somewhere a client could find it without asking, and somewhere you or your team can point to if the question ever comes up.
Will disclosing cost me clients?
The research says disclosure carries a small, consistent trust cost. It also says exposure carries a larger one. If you are using AI in ways a client would object to, the disclosure is not the problem and no wording fixes it. If you are using it for the infrastructure of remembering, most people find that unremarkable once it is explained calmly.
What about AI note-taking during sessions or client meetings?
That is a genuinely different question, involving consent, recording law, retention, and what happens when a client asks for their notes, and the answer varies by jurisdiction and profession. I am not going to answer it in a paragraph. Get it right before you switch anything on.
Sources: O. Schilke and M. Reimann, The transparency dilemma: How AI disclosure erodes trust, OBHDP 188, 2025, thirteen preregistered experiments. Pew Research Center, From political speeches to songs, September 2025, 5,023 US adults. ICF Code of Ethics. ICF AI Coaching Framework and Standards. APA health advisory, November 2025.