EU AI Act Chatbot Disclosure: A Guide for Online Stores
· 8 min read · AskMerra
Since 2 August 2026, a new rule applies to AI chats that talk to people in the EU: people have to be told they are talking to an AI. The rule comes from Article 50 of the EU AI Act, and it covers the chat assistant on your online store just as much as a bank's support bot.
The rule itself is short. The practical questions are not. Does your chat count? Who is responsible for the notice? What should it say, and in which languages? This guide answers those questions for store owners, using the European Commission's own explanations as the source and AskMerra's widget as a worked example.
This is a practical guide, not legal advice. Check your own setup with your legal advisor.
What Article 50 says, in plain words
Article 50(1) of the AI Act says that providers of AI systems meant to interact directly with people must design them so that people are informed they are interacting with an AI system. There is one exception: when this is obvious to a reasonably well-informed, observant and circumspect person, given the circumstances and the context of use.
Article 50(5) adds how: the information must be given "in a clear and distinguishable manner at the latest at the time of the first interaction", and it must meet the applicable accessibility requirements.
The European Commission's FAQ on Article 50 adds three points worth knowing:
- Article 50 applies from 2 August 2026. The only grace period concerns a different obligation, the machine-readable marking of AI-generated content, and only for systems placed on the market before that date (until 2 December 2026). There is no grace period for telling people they are talking to an AI.
- National market surveillance authorities do most of the enforcement.
- Fines can reach 15 million euros or 3% of total worldwide turnover for the previous financial year, and proportionality can be taken into account for small and medium-sized companies.
Source: European Commission, Transparency obligations under Article 50 of the AI Act, last updated 24 July 2026.
Does the chat on your store count?
According to the Commission's guidelines, the obligation applies when four conditions are all met:
- the system qualifies as an AI system;
- it is designed for a genuine two-way exchange with people, not just to collect data or send automated responses;
- the interaction is direct: the AI itself communicates with the person, not through a human intermediary;
- the other side is a natural person, whether a consumer, a professional or another user.
An AI shopping assistant that answers shoppers in a chat window meets all four. It is the textbook case. Here is how other tools on a typical store compare:
| Tool on your store | Covered by the disclosure rule? |
|---|---|
| AI chat assistant answering shoppers | Yes, the clearest case |
| Live chat where your team types every reply | No, no AI is talking |
| AI drafts replies, and a person reviews and sends them | Condition 3 points to no: the AI talks through a human |
| "Customers also bought" blocks and search ranking | No, they work in the background without a direct exchange |
| A button menu with fixed answers and no AI behind it | Possibly not an AI system at all, so ask your vendor |
The last row deserves a check. If a "simple" bot writes its replies with a language model, it is an AI chat.
The Commission's FAQ names chatbots, AI agents and avatars as examples of systems that interact directly with people. If you are looking at agents that act for your shoppers, our guide to AI agents for ecommerce explains the different kinds.
Provider and shop: who does what
The AI Act separates two roles.
- The provider develops an AI system, or has it developed, and places it on the market or puts it into service under its own name or trademark. Article 50(1) is addressed to providers: they must build the disclosure into the system.
- The deployer uses an AI system under its authority, for professional purposes. A store that adds a chat assistant from a vendor is usually a deployer.
So when you use a ready-made assistant, the duty to design the notice sits with the vendor. That does not mean you can ignore it. You control the settings that decide whether the notice works in practice: the assistant's name and avatar, the greeting, pop-up messages and, in many tools, custom CSS that can hide any element on screen. A notice the vendor built and you then hid informs nobody.
If you build your own chatbot, or have an agency build one that you run under your own brand, you may be the provider yourself. Where exactly the line falls for a white-labelled tool is a question for your legal advisor.
What a good disclosure looks like
The rule gives three tests: clear and distinguishable, at the latest at the first interaction, and accessible. On a store, that means:
- Visible before the first answer. The notice sits in the chat window, where the shopper is reading, not only in your terms of service or in a tooltip.
- Plain words. "You are chatting with an AI assistant" works. "Powered by smart technology" does not.
- Real text. Screen readers can read it, and it grows when the shopper zooms in. Text inside an image does neither.
- In the shopper's language. If you sell in six languages, you need the notice in six languages.
- Consistent with the rest of the chat. A human first name, a stock photo of a smiling agent and "let me check with my colleague" all suggest a person, even with a notice above them.
For reference, this is the line AskMerra's widget shows at the top of every conversation, in each interface language:
| Language | Disclosure line |
|---|---|
| English | You are chatting with an AI assistant. It can make mistakes, so check important details. |
| Romanian | Vorbești cu un asistent AI. Poate greși, așa că verifică detaliile importante. |
| Italian | Stai chattando con un assistente AI. Può sbagliare: verifica i dettagli importanti. |
| French | Vous discutez avec un assistant IA. Il peut se tromper : vérifiez les informations importantes. |
| German | Sie chatten mit einem KI-Assistenten. Er kann Fehler machen, prüfen Sie daher wichtige Angaben. |
| Spanish | Estás chateando con un asistente de IA. Puede equivocarse: comprueba los datos importantes. |
The second sentence goes beyond what Article 50 asks. We added it because it sets the right expectation: the assistant is useful, and for anything important, like a delivery date before a wedding, the shopper should double-check.
The "obvious" exception, and why not to rely on it
You do not have to inform people when it is obvious that they are talking to an AI. The benchmark is a reasonably well-informed, observant and circumspect person, and the Commission says the exception should be interpreted restrictively, because it takes transparency away from people.
On a store, that is rarely a safe bet. Modern assistants write fluent, friendly sentences. Many shops give theirs a first name. A shopper on a phone, in a hurry, will not stop to wonder who is typing. A one-line notice costs you nothing. Arguing afterwards that your chat was "obviously" an AI is a discussion you would rather not have with a regulator.
A checklist for store owners
- List every place where AI talks to shoppers: the chat on your site, messaging channels, email auto-replies, any phone line.
- Open each one as a shopper would, on a computer and on a phone, in every language you sell in. Is the AI notice visible before the first answer?
- Check pop-ups and proactive greetings. A bubble that appears next to the chat button before the window opens is often the first thing a shopper reads. Write it so it cannot pass for a person: "Hi, I'm the shop's AI assistant. Looking for something?"
- Review the name and avatar. A name is fine. A name plus a human photo plus no mention of AI is a bad mix. A subtitle such as "AI shopping assistant" settles it.
- Never hide the notice with custom CSS, and tell whoever maintains your theme not to.
- Make the human route visible, and make it clear to the shopper when a person has taken over.
- Link your privacy policy from the chat, and describe the chat in it. Our GDPR checklist for AI chat covers that side.
- Keep a record: dated screenshots of the chat in each language, and your vendor's description of how its disclosure works.
How AskMerra shows the disclosure
Here is how each part of the rule maps to the widget, so you can check it yourself:
- The notice opens every conversation. The disclosure line sits at the top of the chat, above your greeting, so the shopper sees it before the first answer. It is plain text inside the chat window, not an image.
- It speaks the shopper's language. The line is shown in the widget's interface language, which follows your page language, in English, Romanian, Italian, French, German or Spanish.
- It links to your privacy policy. Set the address under Privacy policy URL in the widget's behavior settings. If you leave it empty, the widget links to AskMerra's privacy notice for shoppers.
- "Are you a bot?" gets a straight answer. In all six languages, that question is answered by a fixed rule, without calling the AI model: the assistant says it is an AI shopping assistant and offers to pass the message to a person.
- People are named when they join. On Growth and above, when someone from your team takes over in live chat, the shopper sees their first name and a "joined the chat" message, and the AI stays silent. When the chat ends, a message says the AI assistant is back.
- The default title says what it is. Until you name your assistant, the header reads "Shopping assistant".
The settings that could weaken the notice are in your hands: the assistant's name and avatar, the greeting, the proactive greeting (off by default) and custom CSS on Growth and above. Go through the checklist again whenever you change them.
Get it right from the first day
If you are setting up an assistant now, you can run every check above before a single shopper sees it. Create your account to set up your catalog, your widget and your privacy link, or book a demo and we will go through the disclosure in each of your languages together. The widget documentation describes every setting mentioned here.
Put AskMerra to work on your store
Connect your catalog, start your 14-day free trial and test the assistant in the playground before you go live.
Start free trialKeep reading
- Can Someone Run Up Your AI Chat Bill? How AskMerra Stops ItA public chat widget is an open door. Domain lock, proof-of-work sessions, rate limits, circuit breakers and spending caps keep your AI budget under control.
- AI Chat and GDPR: A Practical Checklist for Online StoresAdding AI chat to an EU online store? A practical GDPR checklist: what the chat collects, EU hosting, retention, deletion requests, the DPA and disclosures.
- AI Agents for Ecommerce: What They Are and Where to StartShopping assistants, support agents, agentic commerce and back-office tools: the four kinds of AI agents for ecommerce, what they need and where to start.