Should you tell customers it's AI answering them?

You put an AI front desk on your website, and the first awkward question isn't from a customer. It's yours: do I tell people?

Short answer: yes, tell them. The longer answer is worth five minutes, because the best-known research on this question appears to say the opposite, and because how you disclose decides whether honesty costs you leads or wins them.

The study that says you shouldn't

Fair fight first. The strongest evidence for hiding your bot is real and peer-reviewed. In 2019, a team of marketing researchers published a field experiment in Marketing Science: more than 6,200 customers of a financial services company got outbound sales calls from either human agents or AI chatbots. The undisclosed chatbots sold as well as proficient human agents and four times better than inexperienced ones. But when customers were told upfront they were about to talk to a bot, purchases dropped by roughly 80 percent. Customers got curt and bought less, judging the same bot less knowledgeable and less empathetic once they knew.

If you stop reading there, the playbook writes itself: shut up and let the bot pass.

Here's why that's the wrong lesson for a trades business in 2026. The experiment was cold-call telemarketing: a bot phoning strangers, passing as human, back when a convincing machine voice was novel enough to feel like a trick. Your setup is a chat window on your own website, opened by the customer, in a year when most people already assume software is on the other end. And the study measures exactly one outcome: the sale on that call. It has nothing to say about the customer who finds out a week later.

Getting caught costs more than disclosing

That customer is the whole argument.

The moment someone discovers the "receptionist" they liked was software coached to pass as a person, every fact it gave them gets re-audited. The price range, the warranty, the "we can be out Wednesday morning": all of it now came from a business whose opening move was a small lie. A front desk exists to build enough confidence to earn a name and a phone number. A discovered fake spends confidence, retroactively, on everything it already said.

And the preference isn't ambiguous. When Qualtrics surveyed consumers in 2023, 88 percent said they want to know when they're dealing with something created by AI, and 41 percent said they always want to know. The same survey found a majority open to AI even in medical and financial matters. Open, as long as nobody hides it.

In two states, part of this is already law

California's Bolstering Online Transparency Act has been in effect since July 2019: using an undeclared bot to push a sale, with intent to mislead people about its artificial identity, is illegal there, and clear disclosure is the explicit safe harbor. Utah's Artificial Intelligence Policy Act took effect in May 2024 and gets to the point faster: when a customer asks a business's AI whether it's AI, the answer has to be honest. (A 2025 amendment narrowed exactly when the duty kicks in, but the ask-and-you-must-tell core is intact.)

Will a two-truck plumbing shop get fined over a chat widget? Probably not this year. The direction is what matters: legislatures keep converging on the same rule your customers already apply. A bot coached to dodge "am I talking to a real person?" is being coached toward, at best, slime, and in a growing number of states, liability.

Disclosure reads as competence, or it doesn't work

Here's the part most advice skips, and it's hiding inside that same 2019 study. The damage from disclosure came from expectations: told "bot," customers assumed they'd been handed the intern before the conversation even started. Which means the disclosure question and the competence question are the same question. An agent that announces itself and then answers badly confirms the suspicion. An agent that announces itself and then answers well breaks it.

So the real work is in the next two messages. "My water heater's leaking from the bottom, is that an emergency?" should get a straight answer: yes, that usually means the tank itself, worth shutting the water off tonight. "Roughly what does a repipe run?" should get the range you approved, and if you never approved one:

That one depends on the layout, so I don't want to guess. Let me take down the details and I'll have Mike confirm the exact price today. What's the best number to reach you?

That reply is disclosure working for you. No human pretense, no fake confidence, and a specific next step with a real person's name in it. Nobody reads that and feels tricked. It reads like a well-run shop, because a shop that set this up is one.

How to disclose without an essay

The practical version, in four rules:

• Label the surface, then let it work. A clear line at the top of the chat does the legal and moral job. The agent doesn't need to reintroduce itself as artificial in every message; it speaks for your business, as "we."
• Answer the direct question honestly, every time. "Am I talking to a person?" gets "No, I'm the AI front desk for Reyes Plumbing. I can answer most questions about the work and pricing, and I can set up a callback with Mike if you'd rather talk to him."
• Keep the human path one message away. The fastest way to make software feel like a trap is to gatekeep the person behind it. Anyone who asks for a callback gets one, no hoops.
• Never let it guess. Disclosure buys you the benefit of the doubt exactly once. A made-up price spends it, and this time the customer knows whose name is on the website.

Where we stand

ServiceFront's agent follows all four, and we'll be straight about what it is: a front desk for your website, answering from a profile of facts you verified, capturing the lead (name, number, the job) and emailing it to you. It doesn't answer your phone, doesn't book your calendar, and doesn't send texts. The calendar and the texts are roadmap; answering your phone isn't. Ask it whether it's AI and it says yes, first try, because we'd rather you win the trust argument than the Turing test.

If you want to hear what honest sounds like for your business, paste your URL at servicefront.ai and ask it the question yourself.