Tell customers whenever AI talks to them directly, shapes a decision about them, creates realistic images or voices, or does work they believe a person is doing. You don't need to announce AI that only helps you draft, proofread or organise work that a person then checks and owns. When unsure, ask: would they feel misled finding out later?
The law sets a floor in a few places. If you sell to customers in the EU, the EU AI Act's transparency rules for chatbots and deepfakes have applied since 2 August 2026. Contracts, platform rules and plain trust do the rest, which is why most disclosure decisions in a small business are judgement calls rather than legal ones.
Four situations where customers expect to be told
These four cover almost every case where hiding AI would feel like a betrayal once discovered.
- AI is talking to them. A website chat, an automated phone line, or emails that go out without a person reading them first. People adjust what they say and how much they trust the answer depending on whether a person is on the other end.
- AI shapes a decision about them. Screening job applicants, setting a price, deciding credit terms, or ranking whose enquiry gets answered first. People have a reasonable interest in knowing how they were judged.
- AI creates something realistic. A product photo showing a finish that doesn't exist yet, a cloned voice, a "customer" in an advert who was never a customer. Realistic synthetic media can mislead even when nobody means it to.
- AI does work they think they're paying a person to do. Translation, copywriting, design or a written report sold as bespoke expert work. If a client would price the job differently knowing a machine did the first 80%, they're owed that knowledge.
Outside these four, AI is a tool like a spell checker or a template. You don't tell customers you used a spreadsheet to calculate their quote, and you don't need to tell them AI tidied the wording of an email a person wrote and sent.
The borderline cases are where the "would they feel misled?" test earns its keep. Three illustrative ones, with verdicts:
- An accountant's year-end letter to a client, drafted by AI from the partner's notes, then edited and signed by the partner. No need to flag it in the letter. The advice is the partner's, and a general statement on the website covers how the firm works.
- A dog groomer's social post presenting "a customer's story" that AI wrote by blending feedback from three different customers. Disclosure doesn't rescue this one. It describes a customer who doesn't exist, so it shouldn't be published at all; quote one real customer, with their permission, instead.
- An AI-generated photo of a smiling "team" on an electrician's About page. Visitors will assume these are the people who'll turn up at their door. Use real photos of the real team, or none.
Notice that two of the three aren't solved by adding a label. When honest disclosure would make the thing pointless, the thing itself is the problem.
Where rules already require it
This is a practical summary, not legal advice. If your situation is borderline, ask a solicitor who knows your sector.
- The EU AI Act, if you sell to customers in the EU. Article 50 requires that people are told when they're interacting with an AI system, unless that's obvious to a reasonably well-informed person. That duty falls mainly on the provider who builds the system, but you're the one deploying it, so check that any chatbot or voice agent you use says so clearly. Deployers must disclose deepfakes, meaning realistic generated or manipulated images, audio or video, with a lighter touch for obviously artistic or satirical work. AI-generated text published to inform the public on matters of public interest must be disclosed unless a person has reviewed it and someone holds editorial responsibility. The information has to be given clearly, at the latest at the first interaction, and the EU has since published guidelines explaining these transparency obligations.
- Platform rules. Some platforms set their own requirements. YouTube requires creators to disclose realistic altered or synthetic content, but not AI used for productivity such as scripts or captions. Check each platform you publish on.
- Fake-review rules. Consumer-protection rules in several major markets treat AI-written reviews and testimonials that don't reflect a real customer's experience as fake reviews. Where the legal line sits for AI testimonials goes through this in detail.
- Your contracts. Some business clients now ask suppliers to disclose or get approval for AI use, much as they would for subcontracting. Read what you've signed, and see how to mention AI use in client contracts and proposals.
- Data protection. If AI tools process customers' personal data, your privacy notice should say so. Decisions made solely by automated means that significantly affect people carry extra rights under data-protection law such as the GDPR. Your data-protection adviser can tell you what your notice needs.
A disclosure decision table for 11 common uses
| What AI does | Tell customers? | Where and how |
|---|---|---|
| Website chatbot answering questions | Yes, always | In the chat's first message, before the customer types |
| AI voice on your phone line | Yes, always | In the greeting, with a way to reach a person |
| Emails drafted by AI, read and sent by staff | Not each time | A general line on your website or in your terms |
| Automatic replies sent with no human check | Yes | A signature line marking the reply as automated |
| AI note-taker on client calls | Yes, and ask first | In the invite and again at the start of the call |
| Machine translation with full human post-editing | Yes | In the quote and your terms, named as its own service |
| Raw machine translation with no human edit | Yes, prominently | In the quote and on the delivered document |
| Realistic AI product or marketing images | Yes, if they could be taken as real | A caption or label; never use them as proofs or product photos |
| AI helping to screen job applicants | Yes | In the job advert and your recruitment privacy notice |
| AI summarising customer feedback internally | No | Your privacy notice covers the data processing |
| AI-drafted website articles a person edits and signs off | Optional | A general statement if you like; the editor answers for the content |
Take a three-chair hair salon working through the table, as an illustration. The salon has an AI voice answering the phone out of hours, writes its social captions with AI, and has started using an AI retouching tool on before-and-after photos. The phone line gets a disclosure in its first sentence: "Hi, you've reached the salon's automated booking assistant. I can book, move or cancel appointments; say 'message' to leave one for the team." The captions need nothing, because a person writes the final version and nothing in them is a factual claim about a customer. The before-and-after photos are the hard one. Smoothing a stray hair in the background is editing; changing the colour result or the skin is showing a finish the salon didn't deliver. The rule the owner settles on: AI retouching on backgrounds only, never on hair or faces, and no label needed because nothing about the result is altered.
Two rows carry the most risk for small businesses: the chatbot and the note-taker. Both involve a customer who hasn't chosen to deal with AI. For the chatbot's opening lines, what to tell customers at the start of a chat has tested wording.
The note-taker has an edge case people forget: the person who joins late. The invite mentioned the note-taker and the host said so at the start, but ten minutes in, the client's finance director joins from another meeting and sees an unfamiliar participant called "Notetaker". Nobody told them. The simple habit is to repeat the line whenever someone new arrives: "Just so you know, we're using an AI note-taker for this call; say if you'd like it off." If they object, pause it and take notes by hand for the rest of the call.
Worked example: a translation agency decides what to tell clients
Take an illustrative eight-person translation agency. It offers three kinds of work: human translation, machine translation followed by full human post-editing, and certified translations for official use. Clients have never been told which jobs use machine translation, and a large client has just sent a supplier questionnaire asking.
The owner works through the table and decides:
- Name the services separately. Quotes now list "human translation" and "machine translation with full human post-editing" as distinct lines. For an illustrative 10,000-word technical manual, the agency quotes $1,400 and five working days for the first, or $900 and three working days for the second. The client chooses knowingly: in this example, it picks the cheaper option for internal documents and human translation for anything customer-facing.
- Point to a standard. The agency's post-editing follows ISO 18587, the international standard for full human post-editing of machine translation output, and its terms say so. That answers "is it just a machine?" before it's asked.
- Keep certified work human-checked, line by line. Certified translations may use machine translation as a first draft, but the certifying translator reads every sentence against the source, and the certificate means exactly that.
- Answer the confidentiality question. The questionnaire also asks where client documents go. The agency names its machine translation engine, confirms client text isn't used for training, and moves one team member off a free online tool. Whether it's safe to put client documents into AI translation tools covers what to check.
The questionnaire now gets a straight answer, and the agency no longer depends on nobody spotting machine translation in a file's metadata.
Telling existing customers about a change
Disclosure is easy for new customers: it goes into the quote, the terms or the chat window from day one. It's harder when you start using AI in a service that existing customers already receive, because they signed up for something that worked differently.
Three steps keep that change from feeling underhand:
- Tell them before, not after. A short email a few weeks ahead of the change, saying what's changing, what stays the same and who still checks the work. "From next month, first drafts of your monthly report will be prepared with AI; your account manager still writes the recommendations and signs off every figure." (A full version of this email follows the list.)
- Offer a choice where it's reasonable. If a client would genuinely prefer the old way, say whether you can keep offering it and at what price. Many won't take it up, but being offered the choice matters.
- Update the paperwork at the same time. Terms, privacy notice and any service descriptions on your website should match what you told them. A mismatch between your email and your terms is the kind of detail that surfaces in a dispute.
Here is that email written out in full, for an illustrative five-person marketing agency whose clients receive a monthly performance report:
Subject: A change to how we prepare your monthly report
Hi [first name],
From the November report onwards, we'll use AI to prepare the first
draft of your monthly report: pulling the figures from your ad and
analytics accounts into our template and writing a plain-English
summary of what moved.
What stays the same: [account manager] still checks every figure
against the source, writes the recommendations, and signs the report
off before it reaches you. Your data stays in our business AI account,
which doesn't use it for training.
What changes for you: the report should arrive two working days
earlier. If you'd prefer we keep preparing it entirely by hand, just
reply and we'll carry on as now.
[Name]Every sentence answers a question a client would otherwise ask: what exactly, who checks, where does my data go, what's in it for me, and can I say no.
Offering a choice means being ready for someone to take it. Say one client replies: "We'd rather our campaign data didn't go into any AI tool." An illustrative answer from the agency:
"Thanks for letting us know, that's no problem. We'll keep preparing your report by hand, exactly as now, and your data won't go into our AI workspace. The report will keep arriving on its usual date rather than two days earlier. If you'd ever like to see a sample AI-prepared report using made-up figures, so you can compare, just say."
It confirms the choice, states the one practical difference and leaves the door open without pressing. Then make sure the choice sticks: flag the account in whatever the team uses to schedule reports, or the next new starter will put that client's data straight into the AI draft.
How customers react depends heavily on the framing. "We've replaced our support team with AI" invites complaints. "AI now handles order-status questions around the clock, so our team has more time for the problems that need a person" describes the same change in terms of what the customer gets.
Wording you can copy
Keep disclosures short, factual and in your normal voice. They should read like information, not a warning label.
WEBSITE (general statement)
We use AI tools to help draft, summarise and organise our work.
A member of our team checks and takes responsibility for everything
we send you.
CHATBOT (first message)
Hi, I'm an automated assistant. I can answer common questions about
[orders, opening hours, pricing]. Type "person" at any time to reach
a member of the team.
AUTOMATED EMAIL (signature line)
This is an automated reply. A member of the team will follow up
by [time] if you need more help.
NOTE-TAKER (calendar invite and start of call)
We'd like to use an AI note-taker on this call so nobody has to
take notes. The recording and notes stay in our business account.
Let us know if you'd rather we didn't.
QUOTE CLAUSE (machine translation with post-editing)
This option uses machine translation followed by full review and
editing by a qualified translator. Choose human translation if you
need every sentence translated from scratch.
IMAGE CAPTION
Illustration created with AI. The finished product may differ
in colour and detail.For a fuller statement to publish on your website, how to write an AI disclosure statement walks through each section.
What happens when customers find out anyway
Customers discover undisclosed AI in predictable ways: a phrase they've seen in other AI text, a chatbot that gives itself away, a file's metadata, a detail that's confidently wrong. What damages trust at that moment is rarely the AI itself. It's the sense of having been kept in the dark.
A realistic way it goes wrong, as an illustration: a small online kitchenware shop gives its website chatbot a friendly persona, "[first name] from the support team", with instructions to sound human. A customer with a warranty problem asks, "Am I talking to a real person?" and the bot, following its persona, replies "Yes, I'm [first name], one of the team here." Two messages later it gives a warranty answer that contradicts the shop's terms. The customer posts both screenshots in a review. The wrong warranty answer was fixable; the claim to be a person was what readers reacted to. The fix took five minutes: the chatbot's first message now says it's an automated assistant, the persona instruction is gone, and "Am I talking to a person?" gets an honest answer and a route to one. If you sell to customers in the EU, the original set-up would also have fallen foul of the chatbot transparency duty.
A disclosure made up front reads as confidence in how you work. The same fact revealed after a customer asks reads as a confession. Worst of all is denying it when asked directly; if someone asks whether AI was involved, say yes and explain how a person checked the work. The AI mistakes that damage customer trust covers this and the other ways AI use goes wrong in front of customers.
A good test before you publish anything: imagine the customer reading an honest description of how it was made. If that description would bother them, either disclose it or change how you make it.
Further reads
- Should You Tell Customers When You Use AI in Your Marketing? — The same question for marketing copy, ads and social posts.
- How to Label AI-Generated Images and Video on Social Media — How to label AI images and video on each social platform.
- Should Freelancers Tell Clients They Use AI? — Disclosure when clients hire you personally for your skills.
- How to Measure Customer Reaction After Introducing AI — Check how customers actually respond once you've told them.
- AI Compliance Checklist for Small Businesses: What Applies to You — Which AI rules apply to your business, beyond disclosure.
- Are AI Meeting Note-Takers Safe for Client Calls? — Consent and data questions before recording client calls.
- How to Map Your Customer Journey and Find Where AI Helps — A journey grid, evidence sources and the signs a step suits AI, walked through with a new puppy owner's first year at a veterinary practice.
- Should You Charge Clients Less When AI Speeds Up Your Work? — What to do about your prices when AI cuts the time a job takes: the rules for hourly billing, when fixed fees are fair, and how to tell clients.
- What to Do When AI Gets Something Wrong With a Customer — The first hour after an AI error reaches a customer: whether to honour what it said, apology wording for email and phone, and how to stop it happening again.
- Do I Need AI in My Business? A Decision Guide for Owners — A decision guide for owners: the three situations where AI is worth it, when waiting is sensible, a ten-question score and a 30-day test.
- Pros and Cons of Using AI in Your Business, With Real Costs — Every pro and con of business AI with a price attached, a first-year ledger for a small law firm, and the costs that never reach an invoice.
- Using AI in a Yoga or Pilates Studio While Keeping Your Voice — Why AI makes every studio sound the same, and the voice sheet, before-and-after edits and teacher-bio method that keep yours recognisably yours.
- 7 AI Mistakes That Put an Accounting Firm's Client Trust at Risk — Seven ways AI use damages trust in an accounting practice, each with how it shows up with clients and the control that prevents it.
- How Business Coaches Use AI Without Losing the Personal Touch — Where AI belongs in a coaching practice and where it doesn't, with a client memory file, recap before-and-afters and voice rules.
- Can an AI Chatbot Book Valuations for Estate Agents Out of Hours? — What an out-of-hours valuation bot needs: live diary access, eight qualifying questions, written diary rules and a list of things it must never say.
- AI Virtual Staging for Property Listings: Costs and Disclosure — Per-photo and per-listing costs of AI virtual staging, the edits a label can't excuse, label wording to copy, and a pre-publish check for every image.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: EU AI Act Article 50 text and the EU's guidelines on transparency obligations; YouTube Help on disclosing altered or synthetic content; ISO 18587 listing (checked September 2026).