Skip to content
disclosura

Law explained

AI Act chatbot disclosure: what your chatbot must say

If customers chat with an AI on your website, WhatsApp, or phone line, Article 50(1) applies to that interaction. The rule: people must be informed they’re interacting with an AI system, unless it’s obvious to a reasonably well-informed person.

Whose duty is it?

Article 50(1) is addressed to the provider: whoever develops the chatbot, or has it developed, and puts it into service under their own name. If you bought a ready-made chatbot, that is its maker. If you built one in-house, or had one built and run it under your business’s name, the Commission’s guidelines (section 2.3) treat you as the provider yourself. Either way you are the face the customer sees, so make sure the notice is there.

What a good disclosure looks like

Clear, upfront, in the user’s language: one plain sentence at the start of every chat session — “You’re chatting with an AI assistant, not a human.” — placed close to the chat field. Art. 50(5): at the latest at first interaction, clearly and distinguishably; the Commission’s guidelines read that as at least once at the start of each interactive session (point 143). A small “AI” badge can stay on screen afterwards, in addition to the sentence (point 37). One prominent notice is likely to be enough in most cases. In riskier settings — children and other vulnerable people, emotional or sensitive conversations, financial, insurance, legal or health advice, complaints handling — reminders are likely to be needed, and in every setting the chatbot must say it is an AI when someone asks (point 40). The Commission’s guidelines (section 3.1.2) list what is not enough on its own: a notice only in your terms, a URL or the documentation; metadata nobody can see; a vague word like “assistant” with no “AI” in it; a general “services on this website use AI”; or a line that only names the technology (“uses LLMs”).

The channels people forget

The duty follows the AI, not the website: WhatsApp bots (put the notice in the first message the bot sends), voice assistants on phone lines (one spoken sentence at the start), Messenger/Instagram auto-agents. The Kit has ready wording for the chat window, for a messaging greeting and for a phone greeting.

The “obvious” exemption

If it’s genuinely obvious you’re an AI, the duty doesn’t apply. The Commission’s guidelines (section 3.2.1) read this restrictively: only where almost no doubt is left, and not where the people who can reach the system include vulnerable people, such as children or older people, for whom the AI nature is less obvious. One of their own examples of a chatbot that does not qualify is a helpdesk chatbot whose replies users may take as human. A one-line notice removes the question entirely. The exemption, examined →

What about human handoff?

If a human takes over the chat, nothing in Art. 50 requires disclosing that — but the reverse switch (human → AI) re-triggers the logic. The guidelines (section 3.1.1) add that where AI replies and human replies are mixed, the AI-generated replies need the disclosure unless a human properly reviewed and sent them. Simplest robust practice: label the AI part clearly, always.

Implementation

This is exactly what our hosted widget does — notice, styling, languages, updates. See it live →

Which of these duties apply to you?

Free, 60 seconds, instant result with exact citations. No email required.

Run the free check

Sources are cited in place; official text: Regulation (EU) 2024/1689. This guide is a technical explanation, not legal advice.