AI Disclosure to Customers: When and How to Say It

AI disclosure to customers means telling people when they are talking to an AI system, or when content they see was generated by one. In the EU the AI Act sets transparency duties for some of these situations, and consumer protection rules already forbid misleading people about who or what they are dealing with. The safe habit is simple: if a reasonable customer would be surprised to learn that a message, a face or a review came from AI, say so in plain words before they find out.

This guide explains when disclosure is expected, what wording works, where to put it and which mistakes damage trust. It is a practical overview for small businesses and teams, not legal advice, so check the current rules for your country and sector. Examples use Ask Mio, and the principles apply to any assistant you use.

Why AI disclosure to customers matters

There are three reasons to disclose, and only one of them is a legal rule.

Law. Some jurisdictions require businesses to tell people when they interact with a machine or see synthetic content. The EU is the clearest example.
Consumer protection. Misleading customers about a material fact, such as whether a human reviewed a message or whether a testimonial is real, can count as an unfair commercial practice.
Trust. Customers forgive an honest “this is our AI assistant” far more easily than a discovery that they were misled. Disclosure also sets expectations, so people know when to ask for a human.

Disclosure does not mean apology. Using AI is normal business practice. The point is to be clear, not to be defensive.

What the EU AI Act says about transparency

The EU Artificial Intelligence Act, Regulation (EU) 2024/1689, contains transparency obligations in Article 50. In summary, and in everyday terms:

Interacting with an AI system. People must be told they are interacting with an AI system, unless that is obvious from the context.
Synthetic content. Providers of systems that generate audio, images, video or text must mark the outputs in a machine-readable way so they can be detected as artificially generated, where technically feasible.
Deepfakes. Those who deploy AI to create or alter image, audio or video that resembles real people, places or events must disclose that the content is artificial.
Text on matters of public interest. AI-generated text published to inform the public on such matters must be disclosed, unless a person has reviewed it and holds editorial responsibility.

The obligations fall on different parties: the provider who builds a system, and the deployer who uses it. A small business that puts a chatbot on its website is normally a deployer. Timelines and details have been subject to amendment proposals, so read the official text and current guidance. Our EU AI Act explainer and the article on AI watermarks and content provenance cover the background. An unofficial but readable Article 50 summary is also available.

Disclosure duties sit alongside data protection law. If an AI system processes personal data, your privacy information must describe it, and rules on automated decisions apply where a system decides something that significantly affects a person. Our AI privacy checklist lists the questions to ask.

When to disclose: a practical guide

Situation	Disclosure	Example wording
Website chatbot answering customers	Expected and usually required where the AI is not obvious	“You are chatting with our AI assistant. A person can join if you ask.”
Voice or phone assistant	Expected at the start of the call	“This is an automated assistant. Say ‘agent’ to reach a person.”
E-mail drafted by AI and reviewed by staff	Usually not required; optional	No label needed, but you remain responsible for accuracy
Fully automated e-mail or message replies	Recommended	“This reply was generated automatically.”
Realistic AI-generated image or video in marketing	Expected where it could be taken for real	“Image generated with AI.”
AI-written articles on public-interest topics	Required unless reviewed with editorial responsibility	“Drafted with AI and edited by our team.”
Testimonials and reviews	Never generate fake ones; label any AI-assisted summary	“Summary of customer feedback, generated with AI.”

Treat the table as a starting point. The right answer depends on your country, your sector and how the content is used. Regulated fields such as health, finance and law often have stricter expectations.

How to word the disclosure

Good disclosure is short, specific and placed where the customer is looking. Four rules help:

Say “AI” or “automated”. Avoid vague wording such as “smart assistant” or “virtual team member” that hides the point.
Say what it does and what it does not. “It answers questions about orders and delivery. It cannot change your payment.”
Offer a person. “Type ‘human’ to reach our team during business hours.”
Give the name a human would not use. Do not give an assistant a human name and photo that suggest a real employee.

Example chat opening: “Hi, I’m Mio, an AI assistant for this shop. I can answer questions about orders, delivery and returns. If you would rather talk to a person, just say so.” The sentence is friendly, honest and gives a way out. Our guide to AI for customer support teams shows how to combine an assistant with human handover.

Where to put it

Disclosure should appear at the point of interaction, not only in a policy page. A footnote nobody reads does not meet the spirit of the rule.

Chat widgets. In the first message and in the widget header.
Voice systems. In the opening line of every call.
Automated e-mails. In a short line near the top or the footer.
Images and video. In a caption or on-image label, and in metadata where your tools support it.
Privacy notice. A fuller description of how AI is used, with data categories and providers, linked from the interface.
About or help pages. A plain explanation of how you use AI and where humans stay in charge.

Keep wording consistent across places, so that a customer who reads your help page finds the same description as the one in your chat.

Content you should never pass off as human

Some uses cross a line even where a specific disclosure rule is unclear.

Fake reviews and testimonials. Generating reviews, ratings or customer quotes is deceptive and illegal in many places. Our guide to responding to reviews with AI shows a legitimate use: drafting replies to real reviews, which you check and send.
Synthetic spokespeople. A lifelike AI face or voice presented as a real employee or customer misleads viewers.
Impersonating real people. Using a real person’s likeness or voice without permission creates legal and reputational risk. Read about the risks in our article on AI deepfake scams.
Claims of human review that did not happen. “Reviewed by our experts” must be true.

Internal practice that supports disclosure

Disclosure works only if the business knows where it uses AI. A few internal habits make that possible:

Keep a list of AI uses. Which tools, for which tasks, with which data, owned by whom.
Write a short AI use policy. Our guide to a company AI use policy gives a structure.
Define human review. Say which outputs a person must read before they go out, and who is accountable.
Test the disclosure. Open your chat widget, call your phone line and read your automated e-mails as a customer would.
Review regularly. Rules, tools and your own use change, so recheck every few months.

An assistant can help with the drafting. Ask for three versions of a chat greeting at different lengths, a one-paragraph description for your help page and a list of places in a customer journey where AI appears. Then review the wording against your actual setup, because the assistant does not know what your systems really do.

Using Mio to draft disclosure text

Ask Mio can draft chat greetings, help-page text and policy paragraphs in Write mode, which is on every plan, including the free one. You describe the real setup, for example “an AI assistant answers order questions and hands over to staff on request”, and Mio produces wording in the length and tone you ask for, in any of its 25-plus languages. You do not choose a model; Mio routes each request to the model it judges best for the task. Memory and projects on paid plans let you keep your wording rules, such as “always mention a human option”, for every future draft.

Mio can help you phrase disclosures, but it cannot tell you what the law requires in your case. For regulated products, public-sector work or large-scale automated decisions, ask a lawyer or your data protection contact to check the final text. Mio’s own data practices are described in its privacy policy, and chats are not used to train models.

Ready-to-adapt wording

These short lines can be adapted to your own tone. Replace the details with what your systems really do, and keep them true.

Chat greeting: “Hello, this is our AI assistant. I can help with orders, delivery and returns, and a colleague can take over whenever you ask.”
Automated e-mail footer: “This message was written by an AI assistant and checked by our team. Reply to reach a person.”
Image caption: “Illustration generated with AI.”
Help page: “We use AI to draft replies and answer common questions. A person reviews anything that affects your account or your payment.”

Read each line from the customer’s side. If it hides the fact that AI is involved, or promises a review that does not happen, rewrite it. Plain statements age better than clever ones, because they stay true when your tools change.

Common mistakes

Burying the notice. A line on a policy page does not inform a person who is chatting.
Human names and photos for bots. They suggest a real person and undermine the disclosure.
Over-claiming. Saying “100 percent accurate” or “reviewed by experts” when it is not true.
No human option. Customers who need a person and cannot reach one lose trust quickly.
Inconsistent wording. The chat says one thing and the privacy notice says another.
Forgetting images. Realistic AI imagery in advertising often needs a label.

Frequently Asked Questions

Do I have to tell customers I use AI?

It depends on how you use it. If customers interact directly with an AI system, EU rules expect you to tell them unless it is obvious. For AI-assisted drafts that staff review, there is usually no specific duty. Whatever the law says, avoid misleading customers about material facts, and check the rules for your country and sector.

What should an AI chatbot disclosure say?

Say plainly that the customer is chatting with an AI assistant, what it can help with and how to reach a person. Place it in the first message and the widget header. For example: “You are chatting with our AI assistant. It can answer questions about orders. Type human to reach our team.”

Do I need to label AI-generated images?

If an image is realistic enough to be taken for a real photo or event, especially one showing people, labelling it is expected and, for deepfakes, required under the EU AI Act. For clearly artistic or illustrative images the duty is lighter, but a short caption such as “image generated with AI” is a safe habit.

Can I use AI to write customer reviews or testimonials?

No. Generating fake reviews or testimonials deceives customers and is prohibited as an unfair commercial practice in many jurisdictions. You can use AI to summarise real feedback, with a clear label, or to draft replies to real reviews that you check and send under your own name.

Is AI disclosure the same as a privacy notice?

No. A disclosure tells people they are dealing with AI or AI-generated content, at the point of interaction. A privacy notice explains how personal data is processed, by whom and on what legal basis. You normally need both, kept consistent, and each should be easy to find and written in plain language.

The Bottom Line

AI disclosure to customers is easiest when you treat honesty as the default: label chatbots, voice systems and realistic synthetic media, offer a human, never fake reviews or reviewers, and keep your wording consistent across chat, e-mail and policy pages. Check the current rules for your country and sector, and ask a lawyer for regulated cases. You can draft your disclosure text on the free plan, and compare options on the Ask Mio pricing page.

AI Disclosure to Customers: When and How to Say You Use AI