Should you tell customers you use AI?

Disclose AI to customers when they’re talking to it, when it made something they’d assume a person made, or when it decides something about them.

Outside those three, it’s mostly a question of trust and manners rather than law. Some owners say nothing because it feels like owning up to a shortcut, and others worry they need a disclaimer on every email. Neither is right.

This is general information, not legal advice. If you work in a regulated profession, or you use AI to make decisions about people, check the detail with a solicitor or your professional body.

Do you have to tell customers you use AI?

Sometimes. There is no general duty to announce that your business uses AI. The rules care about three narrower things: whether a customer could be misled about who or what they are dealing with, whether a machine is making decisions about them, and what happens to their personal data.

The AI disclosure rules that matter to an Irish small business come from the EU AI Act, GDPR and ordinary consumer protection law. All three are aimed at the same thing: people shouldn’t be left with a false impression.

Where AI disclosure is required

In a handful of situations, telling people is a legal requirement, or close enough to one that you should treat it that way:

  • A chatbot or voice assistant that talks to customers. The EU AI Act’s transparency rules say people should be told when they are dealing with an AI system, unless that is already obvious. A chat window on your website counts, and so does an assistant that answers the phone after hours.
  • Realistic images, audio or video that AI made or altered. If it shows a real-looking person, place or event that someone could take as genuine, the Act expects it to be labelled.
  • Decisions about people made by a machine. Under GDPR, a decision made purely by automated means that significantly affects someone (turning down an application, say) brings a right to be told and a right to have a person review it. Keeping a person on the final call keeps you clear of it.
  • Personal data going into an AI tool. If customer details pass through an AI service, your privacy notice should say what the data is used for and who processes it. An accurate notice covers this; you don’t need a banner on every message. There is more in is it safe to put client data into AI tools?
  • Reviews and testimonials. Consumer law already forbids misleading customers, and fake reviews are about as plain an example as there is. AI-written reviews, or an invented customer quote on your website, are a problem with or without a label. Don’t do it.

The AI Act’s transparency duties apply from August 2026. There has been discussion at EU level about adjusting parts of the timetable, so recheck the date if a decision hangs on it. The EU AI Act in plain English covers the rest of the Act.

Telling clients you use AI when the law doesn’t ask you to

Plenty of everyday AI use carries no legal duty to disclose, but customers would still want to know about it. I’d ask one question: if a client found out later, would they feel misled? If the honest answer is yes, tell them now. That usually applies to:

  • Anything personal. A condolence card, a thank-you to a client of twenty years, a reply to a complaint. If it reads as though you sat down and wrote it, either you did, or you rewrote the draft until it became yours.
  • Paid-for judgement. When an accountant, solicitor or consultant uses AI to draft part of a report, a client paying for expertise will reasonably want to know a qualified person checked it. Some professional bodies have issued guidance on AI, so look up what yours says.
  • A made-up persona. Giving the website assistant a human name and a smiling photo, so people think “Sarah” works for you, is where manners shades into misleading.
  • Sensitive information. Someone sharing medical, financial or family details will trust you more if you say up front where AI fits in and what it never sees.

Say you run a small physio clinic in Ballincollig and use AI to turn session notes into a tidy summary for each patient. A disclaimer on every summary would be overkill. One line on the intake form would do: “We use software, including AI, to help write up notes, and your physio reads every summary before it goes to you.”

When you don’t need to say anything

Most of the ways a small business uses AI need no mention at all. You don’t tell customers you used spellcheck or a template, and AI that drafts something you then read, edit and send under your own name is much closer to those than to a bot answering the phone. That covers:

  • drafting an email or quote that you edit and send yourself
  • summarising a meeting for your own notes
  • tidying the wording on your website or a social post
  • sorting your inbox, or reminding you to chase an invoice
  • internal reports nobody outside the business sees

Once you’ve read a draft, changed what needed changing and put your name to it, the work is yours.

AI transparency for a small business: what to actually say

Keep it short and factual. A disclosure that sounds apologetic, or explains too much, makes people more nervous. One sentence, placed where the customer meets the AI, is usually enough. A few you could adapt:

  • Website chat: “This is our automated assistant. It can answer common questions and take bookings, and you can ask for a person at any time.”
  • After-hours phone: “You’re through to our automated assistant. It’ll take your details and someone will ring you back in the morning.”
  • Proposals and reports: “We use AI tools to help with drafting. Every figure and recommendation here has been checked by [name].”
  • Privacy notice: a plain paragraph saying which kinds of AI tool you use, what customer information goes into them, and why.

Two things count for more than the wording: an easy way to reach a person, and someone checking what the AI says. A label on a chatbot doesn’t excuse it quoting a customer the wrong refund policy.

Will telling customers you use AI put them off?

Rarely, if the AI is doing something sensible. People are used to automated booking confirmations and chat windows by now, and a reply in thirty seconds at ten at night beats one at lunchtime tomorrow. What puts people off is finding out afterwards, or feeling that a bot was pretending to be a person.

If you’d like to work out where AI already touches your customers, and where a line of disclosure or a human check belongs, that fits inside the AI Assessment: a fixed-fee look at how your week actually runs, with a plain plan at the end.

Book the AI Assessment

The assessment is credited in full if you go on to a build, and if I can’t find you five hours a week you pay nothing. See how it works.