Your AI answers WhatsApp? The AI Act followed it there.
Article 50(1) attaches to the AI system interacting with people — whatever the channel. Website chat, WhatsApp Business, Messenger, Instagram DMs, a voice agent on your phone number: same duty, different placement.
The duty is the provider’s: the maker of the assistant, or you, if you built it or had it built and run it under your business’s name (Commission guidelines, section 2.3). Fixed, rule-based auto-replies are not AI systems and are outside the rule (section 3.1.1). An assistant that writes its own answers is inside it.
WhatsApp / Messenger / DMs
Put the disclosure in the first message the assistant sends (“Hi! I’m the AI assistant of [Business] — a human colleague can take over, just ask.”). Promise the hand-over only if a person really can take over; otherwise end the greeting after “[Business]”. The guidelines ask for the notice at least once at the start of each interactive session (point 143), so set the greeting to come back when a new conversation starts, not only the first time a customer ever writes. Naming the business matters: the guidelines (section 3.1.1) want AI agents to say both that they are AI and on whose behalf they act. A line in your business profile can help too, but on its own it is not enough: a disclosure only in documentation or a URL does not count (section 3.1.2). The Kit ships ready wording.
Phone / voice AI
Hardest channel — no screen. One spoken sentence at the start: “You’re speaking with [Business]’s AI assistant.” The guidelines (section 3.1.2) give a similar example (“This is an AI-powered assistant”), suggest reminders in longer calls, and say a tone or sound on its own is not enough. On the phone the “obvious it’s AI” exemption is also weaker: a human-sounding voice makes it less obvious (section 3.2.1). Disclose, always.
Human handoff etiquette
When the bot escalates to a human, say so (“Connecting you to a colleague”); when the AI returns, re-disclose. For phone and voice, the guidelines (section 3.1.2) suggest reminders in longer interactions, in particular after an interruption or a change of the AI system’s role. In any channel, the assistant must be built to say it is an AI whenever someone asks, or is likely to be misled or confused about it (section 3.1.2, point 40). The principle behind Art. 50(1) is simply: nobody should wonder whether they were talking to a machine.
In the Kit
The website widget goes in with one line of code; the channel wordings are copy-paste, in all 24 official EU languages (English, Italian and Romanian reviewed by us; the other 21 machine-translated from the English and labelled as such).
Which of these duties apply to you?
Free, 60 seconds, instant result with exact citations. No email required.
Run the free checkSources are cited in place; official text: Regulation (EU) 2024/1689. This guide is a technical explanation, not legal advice.