The EU AI Act and your website chatbot: telling customers they're talking to AI

The EU’s Artificial Intelligence Act was adopted in 2024 and applies in stages. One of the parts that touches almost every business with an AI assistant on its website is the transparency rule in Article 50: people should know when they’re interacting with an AI system.¹ According to the Act’s timetable, Article 50 applies from 2 August 2026.² The Commission has proposed adjustments to parts of the timetable, so check the current status for your case; the practical steps below are sensible either way.

This note explains what the rule asks of a business that deploys a chatbot or AI assistant to talk to customers. It isn’t legal advice, and for anything unusual you should ask a lawyer. For the common case, an assistant answering questions on a website, the steps are short.

What Article 50 says, in plain words

Article 50(1) puts the duty on providers of AI systems intended to interact directly with people: the systems must be designed so that people are informed they’re interacting with an AI system, unless that’s obvious to a reasonably well-informed person from the circumstances.

For a store using Intercom, Zendesk, Tidio, Chatbase or similar, the platform is the provider and the store is the deployer. In practice the two share the job: the platform builds the means of disclosure, and the store configures and shows it. Most platforms already label AI answers in some way; your job is to make sure the label is on and clear.

Other parts of Article 50 deal with AI-generated or manipulated content, such as synthetic images, audio, video and text published to inform the public. A support assistant answering questions about delivery isn’t the main target of those rules, though an assistant that writes marketing content might be.

The information should be given in a clear and distinguishable way, at the latest at the time of the first interaction, and should be accessible.

What it means for a website assistant

Say it’s AI, at the start. A short line in the chat window’s greeting or header: “You’re chatting with our AI assistant. It answers from our website and can pass you to the team.” Most platforms let you set the greeting and the assistant’s name.

Don’t give it a human disguise. A name like “Sophie” with a stock photo and no mention of AI is the case the rule is written for. A name is fine; a name plus “AI assistant” is clearer.

Keep the label when the conversation moves. If the chat hands over to a person, say so: “You’re now chatting with Marc from our team.” Customers should be able to tell which answers came from whom.

Make it accessible. The disclosure should be in the text of the chat, readable by screen readers, not only in an icon or a tooltip.

What it doesn’t require

The transparency rule is not a ban, a licence or a registration. It doesn’t require you to:

• stop using an AI assistant;
• show a long legal notice before every chat;
• disclose which model or vendor you use;
• label every single message, once the conversation clearly started with an AI.

A website support assistant is not, by default, a high-risk system under the Act. The high-risk rules cover specific uses such as recruitment, credit scoring and access to essential services, and come with their own timetable.¹

The other rules that already applied

The AI Act isn’t the only rule that touches a chatbot. Two others were already in force:

• Data protection. If the assistant collects names, emails or order numbers, the GDPR applies as it does to any form. Your privacy notice should mention the chat and the platform that processes it. France’s data protection authority, the CNIL, publishes guidance on AI and personal data that is useful well beyond France.³
• Consumer law. What the assistant says counts as what the business says. A promised refund, a wrong price or an invented policy can be held against you like any other statement on your site. We wrote about the most common case in why chatbots invent return policies.

The second point is where most practical risk sits. Being transparent that an answer came from AI doesn’t make a wrong answer acceptable to the customer.

A short checklist

1. The greeting says the customer is chatting with an AI assistant.
2. The assistant’s name, if it has one, is followed by “AI assistant” or similar.
3. Hand-overs to a person are announced.
4. The disclosure is in text, not only an icon.
5. The privacy notice mentions the chat and the platform.
6. The assistant’s answers are checked against your pages regularly, because what it says, you said.

The last item is the one we help with: Assistant Care asks your assistant what customers ask and checks each answer against your site, every week. If you’re setting one up from scratch, Assistant Setup includes the disclosure in the greeting by default.

Footnotes

1. Regulation (EU) 2024/1689 (Artificial Intelligence Act), eur-lex.europa.eu. ↩ ↩²

2. European Commission, “AI Act” overview and application timeline, digital-strategy.ec.europa.eu. ↩

3. CNIL, “Intelligence artificielle”, cnil.fr. ↩