Your shop's chatbot and the AI Act: what has to be true on 2 August
In five days, on 2 August 2026, Article 50 of the EU AI Act starts to apply. If you have a chatbot on your site, one requirement lands on you: a person has to know they are talking to a machine.
There is no transition period. It covers chatbots that were already deployed.
That sounds trivial, and mostly it is — the awkward part is when and how the notice has to appear. Below is what the requirement actually says, and what it does not say, because that is where most of the confusion lives.
What Article 50(1) actually requires
One obligation: an AI system intended to interact directly with a person must be designed so that the person is informed they are dealing with an AI.
Four conditions have to be met together before it bites:
- the system qualifies as AI under the Regulation,
- it is intended for genuine two-way exchange rather than data collection alone — an ordinary contact form does not qualify,
- the communication runs directly between the AI and the person,
- the counterpart is a natural person.
An ordinary shop chatbot meets all four.
When: from the start of the first interaction. Not after the third message, not in the terms of service, not behind a “learn more” link.
How: clearly, distinguishably, and accessibly — in line with accessibility requirements. The notice has to let someone make an informed decision about whether to have the conversation at all.
The “it’s obvious” exception is narrower than it looks
The requirement does not apply where it is obvious that the person is dealing with an AI. That reads like a convenient escape hatch. It isn’t one.
The yardstick is an average, reasonably well-informed and observant person, and the exception is interpreted restrictively, precisely because the whole provision exists to create transparency. Calling your bot “AI Assistant” probably clears the bar. Giving it a human first name and hoping everyone infers the rest does not.
In practice: if you are wondering whether it’s obvious in your case, it isn’t.
Who the obligation is actually on
Worth being precise here, because “the shop has to” is a shortcut.
Article 50(1) addresses the provider — whoever designs and develops the system. If you use a third-party chatbot, formally it is your vendor who must design it so the notice appears.
Except that in practice it is your shop standing behind that bot, in front of your customers and in front of a regulator. If your vendor doesn’t provide the notice, or lets it be switched off and somebody switched it off, you have an operational problem regardless of whose name is on the legal obligation. Check it, and have an answer ready.
Two dates people conflate
| Date | What it covers |
|---|---|
| 2 August 2026 | The duty to inform a person that they are interacting with an AI (Art. 50(1)). No carve-outs, including for bots deployed earlier. |
| 2 December 2026 | Only machine-readable marking of AI-generated content (Art. 50(2)), and only for systems placed on the market before 2 August 2026. |
The December date does not extend your deadline for disclosing that someone is talking to a bot. It is a separate obligation about marking generated content, and only for older systems.
If someone tells you “we have until December” — they are talking about something else.
The Commission’s guidelines: new, non-binding, and still the thing to read
On 20 July 2026, eight days ago, the European Commission adopted guidelines filling in Article 50’s fairly open-ended language.
They are formally non-binding. In practice they will be the primary reference for national authorities interpreting the provision, so treating them as optional is a bad bet.
One point worth remembering: automatic AI labels applied by platforms do not discharge an operator’s own disclosure duty. They can supplement it; they cannot replace it.
What’s at stake
Breaching Article 50 carries penalties of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher.
Realistically, a small shop is not getting a €15 million fine. But this is not the provision to test a regulator’s patience on, particularly when compliance costs you one sentence of interface copy.
What to do this week
A list you can work through in fifteen minutes:
- Open your own chatbot as a customer would. Is the AI notice visible on first contact? Not in the terms — in the chat window.
- Check whether it can be turned off. If it can, check whether anyone has.
- Look at the name and avatar. A human first name with no AI marker anywhere is the single most common problem out there.
- Do the same on a phone and with a screen reader. The requirement talks about accessibility, not merely about the text existing.
- Ask your vendor in writing how they satisfy Article 50(1). Keep the answer in your inbox.
Where we stand
So you know where I’m arguing from: in Chatmerce, the AI notice is built into every chat surface and cannot be removed. Pro and Enterprise plans can change its wording; they cannot switch off the fact of it. That was the design before this deadline got interesting.
I’m not writing this to say “we’re fine, buy from us.” I’m writing it because it is the only sound answer to “does your bot comply?”: a notice that is on by default and can’t be turned off, rather than a setting somebody forgets.
Want to check your own bot, whoever built it? We put together a two-minute Article 50 check — five questions and you know what to fix.
This is not legal advice. It describes how we read the provision and what follows from it for our product. If your situation is non-standard — you modify or rebrand somebody else’s AI system, for instance — talk to a lawyer, because at that point “who is the provider” stops being obvious.
More on our approach: AI Act and security.