In this guide
What Article 50 says
Regulation (EU) 2024/1689, the Artificial Intelligence Act, sets out its transparency obligations in Article 50. The first paragraph is the one that applies to an assistant answering on WhatsApp:
“Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use.”
Article 50(1)
And paragraph 5 says how and when:
“The information referred to in paragraphs 1 to 4 shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure. The information shall conform to the applicable accessibility requirements.”
Article 50(5)
In other words: if the person writing to you could think they are talking to a person, you have to tell them they aren’t, and tell them from the start. It applies from 2 August 2026 (Article 113), and the digital omnibus that amended the Act in July, Regulation (EU) 2026/1744, changed neither of these two paragraphs.
What it means on WhatsApp
A WhatsApp chat is exactly the case the article has in mind: the customer writes as they would to a person and, without a notice, has no way of knowing who is answering. The “unless this is obvious” exception doesn’t help here, because on WhatsApp a well-written reply looks like a person’s.
On 20 July 2026 the European Commission published guidelines on Article 50. They are not binding, but they show how the Commission reads it, and almost everything below comes from them.
In the first message
The notice goes in the assistant’s first reply, not in a message that arrives later or behind a link. The guidelines say it should, at a minimum, “be provided once at the start of an interactive session” (paragraph 143).
What it is and who it speaks for
“I’m a bot” isn’t enough. Under the guidelines, the customer has to know that it is an AI and on whose behalf it is acting (paragraph 31). That’s why the sentence that works names the agency: “I’m the AI assistant of Viajes España”.
If they ask, the truth
If a customer asks “are you a person?”, the assistant has to say it is an AI, in the middle of the conversation too (paragraph 40). And it shouldn’t introduce itself with a name and a photo that make it look like someone from the agency (paragraph 38).
When who answers changes
If someone from the agency takes over the conversation and then hands it back to the assistant, it’s worth saying it again: the guidelines recommend a reminder when the system’s role changes during the conversation (paragraph 37). And it’s good practice to repeat it when the customer writes again on another day.
In their language, and in writing
A notice that can’t be understood doesn’t inform: if the customer writes in English, the notice goes in English. And if the assistant replies with voice notes, the notice is best written too, so it can be seen without pressing play.
A first message that complies
The first line says what it is and whose it is. The rest answers what the customer asked.
No legal paragraph, and no welcome message that delays the answer. The notice goes first and on its own line, so it’s seen even if the customer only reads the beginning.
Underneath, in small print, a button to the privacy information. That’s another obligation, the GDPR’s, and it’s handled in the same message.
The 12 April departure is €2,450 per person, with flights and hotels included. How many of you would be travelling?We handle your data to help you21:14Privacy
What doesn’t count
Who has to comply
Paragraph 1 binds the provider: whoever develops the system, or has it developed, and places it on the market or puts it into service “under its own name or trademark” (Article 3, point 3). The agency using it is the deployer (point 4). In practice, it depends on how you have your assistant:
The notice should come built in
The provider is the company that gives you the assistant, and the notice is part of the product. Before signing up, ask how it gives the notice and check it in a test chat.
The provider may be you
If the agency has a custom assistant built and puts it into service under its own name, it may become the provider itself, and the notice its responsibility.
Either way, the agency is the one the customer sees. The practical step is to check that the notice appears: in the first message, with the agency’s name and in the customer’s language.
And if there’s no notice?
The Act sets the ceiling for fines for breaching Article 50: up to €15 million or 3% of worldwide annual turnover for the previous year and, for SMEs, whichever of the two is lower (Article 99(4) and (6)). Each country sets its own penalties within that ceiling. In Spain, the bill that was meant to do so lapsed when Parliament was dissolved (Royal Decree 806/2026, BOE of 6 October), but the duty to give notice doesn’t depend on that law: the EU Act applies directly.
The good news is that complying takes one sentence.
What comes next
The same article has another obligation, in paragraph 2: content generated by an AI (text, audio, images or video) has to be marked in a machine-readable format, so that it can be detected as artificial. That’s for the provider to do and changes nothing in the chat: the customer doesn’t see the mark. For systems already on the market before 2 August 2026, the digital omnibus gives until 2 December 2026.
How Tramoa does it
In Tramoa, the notice doesn’t depend on what the AI writes: the code adds it in front of the first message, with the agency’s name and in the customer’s language. It appears again if the customer writes after 24 hours without talking and every time someone from the agency hands the conversation back to the assistant, with a different sentence so it doesn’t repeat itself. If the customer asks, the assistant says it’s an AI, and hands them over to a person whenever they want.
Sources
- Regulation (EU) 2024/1689, the Artificial Intelligence Act: Articles 3, 50, 99 and 113 (EUR-Lex, in English).
- Regulation (EU) 2026/1744 of 8 July 2026 (Digital Omnibus on AI), OJ of 24 July 2026.
- Commission Guidelines on the transparency obligations under Article 50, C(2026) 5054, of 20 July 2026. They are not binding, and the Commission will formally adopt them once all language versions are available.
- Royal Decree 806/2026 of 5 October, dissolving the Spanish Congress and Senate (BOE, in Spanish).
This guide summarises the rules as published on 28 October 2026. It is not legal advice.