Your customers are uneasy about AI. Talk to them like it.
Most teams that start using AI think hard about the tools and very little about the people on the other end. Clients, patients, parents and residents notice anyway. A reply arrives faster than usual and sounds a little too smooth, or a summary of their own meeting comes back with a detail they never said. Most of them won't ask about it. They just trust the organization slightly less and keep that to themselves.
The unease is reasonable.
Wary customers usually have their reasons. Many of them have already had a bad experience: a chatbot that couldn't hand them to a person, an automated email that got their name wrong, a support exchange that went in circles. They have also heard plenty about data being used in ways nobody agreed to. When they learn that an organization they deal with uses AI, those are the pictures that come to mind.
That means the default reading of your silence is not neutral. If a family finds out from a footnote, or from a strange sentence in an email, that the school district's office drafts replies with AI, they will fill in the gaps with the worst version. The same use, explained up front in two plain sentences, usually lands as unremarkable.
Say where it is, specifically.
The common mistake is a single vague line on the website: "We use AI to serve you better." It tells the reader nothing they can check, and it reads like the sort of sentence written to avoid saying anything. People are more reassured by detail than by tone.
So name the actual uses. A clinic might say that AI transcribes and drafts visit notes, which the clinician reviews and signs before they go into the record. An accounting firm might say that it sorts incoming documents and drafts first-pass reconciliations, and that a staff accountant checks every figure. A town office might say that it suggests answers to common permit questions, and that a clerk reads each reply before it is sent. Each of those tells the reader what the tool touches and who stands behind the result.
People are reassured by detail they can check, far more than by a warm sentence about serving them better.
It is just as useful to say where AI is not used. If nobody's eligibility, diagnosis, grade or bill is decided by a system without a person looking at it, write that down. For many readers, that sentence answers the question they were actually worried about.
Promise a person, and keep the promise.
The strongest thing you can tell an uneasy customer is that they can reach a human and that a human checks the work that matters to them. That promise only helps if it holds. If the statement says a person reviews every message and a client later receives an obvious unreviewed draft, the statement has made things worse.
Before writing anything public, check the promise against how the work runs today. Which outputs does a person read before they leave the building, and which ones go out on their own? Where that answer is uncomfortable, change the workflow first and the wording second. A nonprofit that promises donors a personal reply should make sure its drafted thank-you notes sit in a queue someone clears, and that the queue has a named owner.
Let the people at the front desk answer it.
Customers rarely read a policy page. They ask whoever is in front of them, which is usually the receptionist, the account manager, the person on the phone or the teacher at pickup. If those people don't know what the organization uses AI for, they will guess, shrug or say something that contradicts the website.
Give them the same short statement and a minute to talk it through at a team meeting. Agree on a plain answer to the three questions they will hear most often: do you use AI on my information, does a person check it, and can I ask you not to? Staff who can answer calmly, in their own words, do more for trust than any page.
That last question deserves a real answer. Some organizations can offer a person-only path for clients who ask, at least for certain kinds of work. Others can't, and should say so politely and explain what safeguards apply instead. Both are better than a front desk that has never heard the question.
Keep it current.
A statement written in the spring describes the tools the team used in the spring. As new uses arrive, the page drifts out of date, and an out-of-date disclosure is close to no disclosure at all. Put a review date on it and make one person responsible for updating it whenever the team adds or retires a use. When a customer raises a concern, write it down and look at the list at each review, because those notes show where the wording is not doing its job.
A "how we use AI" statement you can adapt.
Keep it to one short page or a section of your About page, written in the same voice as the rest of your site. Fill in each part from what your team actually does today.
- What we use it for: two to five specific tasks, each in one sentence. "AI drafts replies to routine scheduling requests."
- Who checks it: for each task, the role that reviews the result before it reaches you, or a plain note that it doesn't. "A coordinator reads every reply before it is sent."
- What it never decides: the decisions that always rest with a person. "No application, bill or appointment is approved or refused by AI."
- Your information: what the tools can see, and that it is not used to train outside systems, if that is true for your tools and settings. Confirm it with your vendor before you publish.
- How to reach a person: a phone number or email that goes to a human, and whether you can offer a person-only path on request.
- When this was last reviewed: a date, and the role that keeps the page up to date.
Read the draft aloud to someone outside the team before publishing it. If they come away knowing what AI touches and who to call, it is ready.
