Do You Have to Tell Customers They're Talking to AI?

Part of our complete guide: How to Use AI in a Small Business.

We get asked this by businesses that are about to put an assistant on their phone or their website for the first time, usually a week before launch, and usually after they have already heard three different confident answers.

1. Why This Is Suddenly a Real Question

Five years ago nobody asked, because nobody was fooled. An automated phone system announced itself by being obviously automated. What has changed is that an assistant can now hold a short, ordinary conversation well enough that a caller is genuinely unsure, and that uncertainty is the whole issue.

If you want to feel how thin the line has become, our own AI or Human page is a short game built around exactly that: short passages, one guess, and most people do worse than they expect. That is the environment your customers are now making judgements in.

2. What We Will Not Tell You

We are not going to tell you what the law requires, and you should be wary of anyone who answers that casually. Disclosure rules are moving, they differ from place to place, they differ by channel — a recorded phone call, a text message and a website chat are not the same thing — and some industries carry their own requirements on top of the general ones.

So the compliance half of this is a question for whoever handles compliance for your business, answered against your actual channels before you launch rather than after. What we will say is that the direction of travel is one way, and that building disclosure in now is far easier than retrofitting it into a flow that was designed to conceal it. That is a practical argument, not a legal one.

3. Customers Often Work It Out Anyway

The giveaway is almost never the voice. It is the behaviour. It never interrupts. It never says "hang on" or loses its place. It answers instantly, every time, with the same even tone, and it repeats a phrase in the same shape it used two minutes ago. People notice the pattern long before they notice the audio.

Which means trying to pass as a person is a bet you win most days and lose eventually, and the loss is not distributed evenly. The damage is concentrated in the moment of discovery, because what the customer learns is not that you use software — everybody uses software — but that you preferred they did not know. That is a much more expensive thing to teach somebody.

4. Disclosure Costs Less Than Owners Expect

The fear is that callers will hang up. In practice what loses a caller is an assistant that cannot help them, not one that says what it is. Someone who is told plainly and then gets a straight answer in twenty seconds, at nine in the evening, has had a good experience and will say so. Someone who is not told and gets stonewalled has had a bad one twice over.

It also sets the right expectation, which quietly makes the assistant work better. A caller who knows what they are talking to asks shorter, clearer questions and is far more willing to be handed to a person. Pretending otherwise invites them to test it, and the test is usually something the assistant was never built for.

5. How to Say It Without Killing the Conversation

One short clause, in plain words, at the start — then immediately be useful. Something close to: "You're through to the automated assistant for Smith Plumbing, and I can book you in or get you to someone." That is it. No apology, no paragraph, and no buried legal line nobody listens to.

Three things make the difference between that landing well and landing badly. Offer the route to a person in the same breath rather than keeping it in reserve. Honour it instantly when it is asked for, because a disclosure followed by an assistant that will not let go is worse than no disclosure at all. And do not give it a human first name with a backstory it will defend if pushed — that is the version of this that ends up quoted back at you.

6. The Two Places Hiding It Is Indefensible

The first is a direct question. If a customer asks whether they are speaking to a person, the answer is a straight no, every time, with no deflection and no clever phrasing. There is no version of dodging that question that reads as anything other than what it is, and it is the single most likely thing to end up on a screenshot.

The second is any conversation where the customer is deciding what to say. Health details, money, a legal matter, a safety problem, something about their family — in all of those, who is listening changes what a reasonable person chooses to share. Taking that choice away from them is the real harm here, and it is the reason the rules are heading where they are heading. Our guides to an AI receptionist versus an answering service and AI appointment scheduling both draw the same line in their own contexts.

How to Start

Write the sentence before you write anything else, and read it out loud. If it sounds like a disclaimer you are hiding behind, it is too long. Then decide the handoff: what the assistant says, how fast a person actually gets on the line, and what happens to the call at six in the evening when nobody will. Those two decisions settle most of this.

Then test it the way a customer would, by asking it directly whether it is a person and seeing what it does. That one question tells you more about whether your assistant is honest than any amount of reading, and it is worth asking again every time somebody changes the prompt. For what else to settle before launch, our guide to training AI on your own business documents covers the other half: making sure that when it does answer, the answer is actually yours.

About Lexington AI Solutions: We are a veteran-owned AI automation consultancy based in Lexington, KY, serving businesses throughout the Bluegrass region. Learn more about our team.