Maybe you have a chat on your website. It answers questions while you sleep, books appointments for you, and catches enquiries you would otherwise have had to handle yourself. Or maybe you’re thinking about adding one. Convenient, inexpensive, and customers get help straight away.
Then you hear something about the AI Act, the EU’s rulebook for artificial intelligence. And you wonder: does this apply to me?
If the chat answers using AI, it probably does, and the fix itself takes a few minutes.
What the law actually says
Since 2 August 2026, the rule has applied across the whole EU. If a visitor on your website is talking to an AI, it has to be clear that it’s an AI and not a person. The requirement sits in Article 50 of the AI Act, EU Regulation 2024/1689. Article 50(5) adds that the information must be given in a clear and distinguishable manner, at the latest at the time of the first interaction.
That’s the whole thing. For an ordinary website chat, the fix is usually simple. In most cases it’s enough to add a clear line of text right in the chat, so the visitor knows what they’re talking to.
There is one exception in the text: the disclosure isn’t needed where it is already obvious that you’re dealing with an AI. But the European Commission states in its questions and answers on Article 50 that this exception should be interpreted restrictively. Assume it doesn’t cover you.
Picture a customer typing in the chat at ten in the evening. She thinks she’s talking to you, or to someone at the company. In reality it’s an AI answering. The rule says the customer has the right to know that from the start. Not buried in the small print, but visible the moment the conversation begins.
Does it apply if you bought the chat ready-made?
Here’s a detail that often gets reported sloppily, and it’s worth getting right.
Article 50(1) is formally addressed to the provider — whoever developed the AI system and placed it on the market. It’s the provider who must build the system so that users are told they’re talking to an AI. If you bought a ready-made chat solution and dropped it onto your site, you’re generally not the provider but what the Regulation calls a deployer. The roles are defined in Article 3 and responsibility along the value chain is allocated in Article 25, and they are two different things.
So can you sit back? No, for three reasons.
The information still has to reach your visitor. The provider may have built the feature, but you’re the one configuring the chat, choosing what it’s called and deciding how it introduces itself. If the disclosure is switched off, designed away, or toned down past recognition, it’s your website showing an AI that passes itself off as a person. Checking that the line is enabled and visible is your job, not your vendor’s.
If you had the chat built for you, you are the provider. The definition in Article 3(3) covers not just those who develop an AI system themselves, but also those who have one developed and place it on the market under their own name or trademark. If you commissioned a bespoke chat and put your company name on it, the obligation lands directly on you.
And it’s your customer who gets misled. No customer is going to care who formally carried the obligation. It’s your brand on the page.
What “disclosing” actually means in practice
So how do you meet the requirement? It’s less of a hassle than you’d think.
Add a clear line at the start of every chat. Something like “You’re chatting with an AI assistant.” It should show up the moment the conversation starts, and it should be easy to understand. One sentence is enough.
On 20 July 2026, just two weeks before the rules took effect, the Commission published its final guidelines on the transparency obligations. They are clear about what doesn’t work: a reference buried in your terms of service, a generic label such as “assistant”, or information held only in metadata is not enough. The disclosure belongs where the conversation happens.
So what isn’t enough in practice? A button that just says “Chat” won’t do. Giving the bot a human name like “Anna” without explaining that Anna is an AI won’t do either. In fact, the more your bot sounds and behaves like a person, the more important the clarity becomes. The whole point is that the customer shouldn’t be misled into thinking they’re talking to a human.
Do you have a chat that calls or speaks with a voice? Then the same applies, just audibly. It should be said at the start of the conversation.
Here’s what to do:
- Check whether your chat uses AI or automated replies.
- Add clear text at the start of the chat, for example “You’re chatting with an AI assistant.”
- Show the text immediately, at the latest when the first message is sent.
- Avoid human names without explanation, like “Anna,” if it’s really an AI answering.
- Save a screenshot of how the information is shown, so you can demonstrate what you’ve done if anyone asks.
What this is NOT
There’s a lot of talk about the AI Act, and some of it creates more confusion than clarity. So let’s clear up what this rule is not about.
It’s about chats that talk to people. It’s not about having to label every image or piece of text that AI helped create.
There are separate rules on AI-generated content. Under Article 50(4), anyone deploying an AI system to create deepfakes must disclose that the content has been artificially generated or manipulated, and the same goes for AI-generated text published to inform the public on matters of public interest. Those rules work differently and rarely catch an ordinary business website. The Commission has also produced a code of practice on marking and labelling AI-generated content for those affected.
Through the spring of 2026 there was genuine uncertainty about which parts of the AI Act would be postponed. That question is now settled. The EU’s digital omnibus pushed the high-risk requirements back to December 2027, but left the Article 50 transparency obligations untouched. Law firm Goodwin noted on 3 August 2026 that the Article 50 obligations became generally applicable, and enforceable by national competent authorities, on 2 August 2026 exactly as originally scheduled.
There is one transitional exception, and it doesn’t cover the chat rule. The Swedish Post and Telecom Authority notes that AI systems already on the market before 2 August 2026 have until 2 December 2026 to meet the Article 50(2) requirement on machine-readable marking of AI-generated content. The duty to say that a chat is an AI is not covered by that transition.
In Sweden, supervision is for now allocated through government assignments while complementary national legislation is finalised. The Post and Telecom Authority is the common contact point and market surveillance authority for several areas, and the Swedish Authority for Privacy Protection has been made a market surveillance authority, particularly for AI systems used in law enforcement. Finansinspektionen has a role too. If you operate in another EU country, your national authority will differ.
What happens if you ignore it?
There are penalties tied to the transparency requirements, and on paper they’re large. Under Article 99(4) of the Regulation, they can reach €15 million or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher. For small and medium-sized enterprises, including start-ups, Article 99(6) applies instead: the fine is capped at whichever of the amount and the percentage is lower.
But keep it in proportion. A sole trader with a chat on their site is not what Brussels is chasing. The realistic point is a different one: the fix is almost free and takes a few minutes. The risk, however small, simply isn’t worth carrying when the solution is this easy.
And there’s a better reason than fines. Being open about the fact that a customer is talking to an AI builds trust. People don’t like being misled. A clear little line saying “this is an AI assistant” makes you look honest, not like someone trying to hide something.
We’re a web agency, not lawyers. This article is a practical summary of how the rule works for an ordinary business website, not legal advice. If a larger or more complex AI system is involved, check with a qualified lawyer.
Primary sources: AI Act Article 50 · Article 99 on penalties · Commission guidelines on the transparency obligations (20 July 2026) · PTS on the AI Act
How we’d check your chat
If you already have a chat, we open it the way a new visitor would, on desktop and mobile, and go through what Article 50 and the Commission’s guidelines require: that it’s clear it’s an AI before the first reply, that this is stated in the chat itself and not only in the terms, and that the bot doesn’t carry a human name without explanation. You get the screenshots, the same ones you should keep yourself.
If you’re getting a chat, we build the disclosure in from the start as part of our AI integrations. And if you sell to consumers, the chat isn’t the only new requirement this year: the withdrawal button has been law since 19 June, and if you run a webshop, the Accessibility Act may apply to you too.