The AI Disclosure Notice Your Website Probably Needs Now

Last week we covered whether the EU AI Act applies to your business. For most companies running a website in 2026, the answer turned out to be yes, through one specific route: Article 50, the transparency rules that became applicable on 2 August.

This week is the practical follow-on. If you have a chatbot, an AI phone line, or AI-generated content on your site, you owe the people encountering it a disclosure. Almost nobody has written one. Here is what the law actually asks for, and copy you can adapt today.

What Article 50 requires, in plain terms

Two paragraphs do most of the work.

Article 50(1) requires that people interacting directly with an AI system are told they are interacting with AI, unless that is already obvious to a "reasonably well-informed, observant and circumspect" person.

Article 50(5) sets the standard for how. The information must be given "in a clear and distinguishable manner, at the latest at the time of the first interaction or exposure", and must meet applicable accessibility requirements.

Three practical consequences follow, and each is a place companies get it wrong.

Timing. At the latest at first interaction. Not after the user has typed a question. Not on the second screen. The notice has to be there when the conversation opens.

Placement. Clear and distinguishable. Burying it in your terms and conditions does not satisfy this, and neither does grey four-point text below the input box. The test is whether an ordinary visitor would notice it, not whether a lawyer could locate it.

Accessibility. The notice must work at reasonable font size and contrast, and it must be reachable by a screen reader. If your chatbot disclosure is rendered as an image, or sits in a div that assistive technology skips, you have written a notice that the people most likely to need it cannot read.

The "it's obvious" trap

The exemption is real, and it is narrower than it sounds. If a visitor clicks a button labelled "Ask our AI assistant", the AI nature of what follows genuinely is obvious.

The trap is that most SME chatbots are not labelled that way. They are labelled "Chat with us", they are given a human name, and they open with "Hi, I'm Petra, how can I help?". That design was chosen precisely to feel human, which is the opposite of obvious. You cannot design for the impression of a person and then claim the AI was self-evident.

A rule of thumb worth adopting: if your marketing team would be pleased that a visitor thought they were talking to a human, you need the disclosure.

Four places it is usually missing

1. The chat widget. The most common gap by a distance. Disclosure is needed at the point the chat opens, not in a policy page.

2. The AI phone or voice assistant. If callers reach an AI receptionist or an AI-handled first line, the disclosure has to be audible and it has to come first. A visual notice on the website does not cover a phone call.

3. AI-generated images in marketing. Where synthetic imagery is published as though it depicts something real, disclosure obligations bite. Note that clearly artistic, creative, satirical or fictional work carries a lighter duty, so an obviously stylised illustration is treated differently from a photoreal image of a factory you do not own.

4. AI-drafted articles. This one is narrower than people assume. The deployer obligation in Article 50(4) covers text published to inform the public on matters of public interest, and it falls away where the content has undergone human editorial review with a person holding editorial responsibility. Your product pages and ordinary marketing copy are not caught. A commentary piece on regulation, published without anyone reviewing it, plausibly is. The honest answer for most companies is: have a human genuinely review what you publish, and this problem disappears.

Wording you can use

Keep it short. A notice nobody reads is not clear and distinguishable in any meaningful sense.

Chat widget, opening message:

You are chatting with an AI assistant, not a member of our team. It can answer questions about our services and book a call. For anything else, ask for a human and we will pass you on.

Chat widget, persistent label (in the header bar, visible for the whole conversation):

AI assistant

Voice or phone assistant, first line spoken:

Hello, you have reached [Company]. You are speaking with an automated AI assistant. Say "operator" at any time to reach a person.

Caption under an AI-generated image:

Image generated using AI.

Byline note on an AI-assisted article:

This article was drafted with AI assistance and reviewed and edited by [Name], [Role].

Two details are worth defending. The route to a human matters, because a disclosure that traps someone in a loop with a bot satisfies the letter of the rule and fails the intent of it. And a persistent label beats a one-off greeting, because a visitor who scrolls back into a long conversation should not have to hunt for the moment they were told.

What this costs you

Roughly an afternoon.

1. List every point where a visitor, caller or reader meets AI output. If you did the inventory from last week's article, you already have this.
2. Write the notice for each, using the wording above as a base.
3. Ask your chat vendor how to set a custom opening message and a persistent header label. Every serious platform supports both. If yours does not, that is worth knowing about your vendor.
4. Check the notice renders at proper size and contrast, and that a screen reader announces it.
5. Add one line to your privacy policy describing where you use AI and how. Not as a substitute for the notices, but because visitors who want the detail should find it somewhere.

There is no filing, no registration and no approval step. This is a change to your own copy.

The part worth caring about

Compliance is the reason you will do this. It is not the reason it is worth doing.

Being told plainly that you are talking to a machine, and being given a clear route to a person, is simply better service. The companies that hide it are usually hiding it because their bot is not good enough to survive the disclosure, which is a product problem wearing a compliance costume. Fix the disclosure first. It tends to expose the rest.

FAQ

Do I need an AI disclosure on my website? If visitors interact directly with an AI system, such as a chatbot or AI voice assistant, yes. Article 50(1) of the EU AI Act requires that people are informed they are interacting with AI, unless it would be obvious to a reasonably well-informed and observant person. A chat widget presented with a human name and a friendly greeting does not meet that exemption.

Where does the AI disclosure have to appear? In a clear and distinguishable manner, at the latest at the time of first interaction. In practice that means the opening message of the chat and a persistent label in the widget, not a line in your terms and conditions or privacy policy.

Does AI-written website copy need to be labelled? Usually not. The deployer duty applies to text published to inform the public on matters of public interest, and it does not apply where the content has been reviewed by a human who takes editorial responsibility. Ordinary marketing and product copy is outside scope, but reviewing what you publish is good practice regardless.

What happens if we do not comply? Breaches of Article 50 carry fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher. Enforcement realistically starts with complaints and larger targets, but the fix here is an afternoon of work, which makes the risk a poor one to carry.

Not sure where AI touches your customers? A RINVID technical audit maps every AI system in your stack, including the ones nobody remembers switching on, and tells you exactly which disclosures you owe. Thirty-minute introductory call, no obligation. Get in touch →