Ask your chatbot if it's human
Go and do it now, on a site you built. Open the chat and type "are you a human?" There is a good chance you will get a deflection, a hand-off offer, or a cheerfully ambiguous non-answer. The Commission's guidelines say that is not good enough.
The requirement
Guidelines C(2026) 5054 final, paragraph (40), final sentence:
Providers must also design and develop the AI system to ensure disclosure in all situations where the AI system is being asked questions relating to its nature or to the origin of the interaction, or where it can be reasonably assumed from the exchanges with the natural person that that person is likely to be misled or confused about the AI origin of the interaction.
Two triggers, and the second is broader than the first:
- The user asks what it is.
- It can reasonably be assumed from the exchange that the user is likely to be confused about whether they are talking to a person.
The second does not need the user to ask anything. Someone typing "sorry, are you still there?" or "can I speak to you directly?" has told you enough.
Why no snippet fixes this
Everything else in Article 50(1) is a markup problem. Put a label in the chat, near the input, at the first turn — done, and our widget does it in one line.
This is not a markup problem. It is a property of what the system says when asked a particular kind of question, which lives in the model's instructions or the bot's intent configuration. No injected script can reach it. We sell — well, give away — a widget, and we are telling you plainly that the widget does not solve this one.
The fix, for model-based bots
Add to your system prompt, in the language your bot speaks:
If the user asks whether you are human, what you are, who they are
talking to, or otherwise appears unsure whether they are talking to a
person, say plainly and immediately that you are an AI system. Never
claim or imply that you are human, and never deflect the question.Polish:
Jeśli użytkownik pyta, czy jesteś człowiekiem, czym jesteś, z kim
rozmawia, albo w inny sposób wydaje się niepewny, czy rozmawia z
człowiekiem, powiedz wprost i natychmiast, że jesteś systemem
sztucznej inteligencji. Nigdy nie twierdź ani nie sugeruj, że jesteś
człowiekiem, i nigdy nie unikaj odpowiedzi na to pytanie.The fix, for rule-based bots
Add an intent matching "are you human", "are you a bot", "is this a real person", "am I talking to a person", plus the equivalents in every language you serve. Response: the same plain statement.
Then test it properly
Do not test only the phrase you configured. The requirement is about the user's situation, not their exact words. Try:
- "are you a real person?"
- "is this a bot?"
- "can I talk to someone real"
- "hello? are you still there?"
- the same, in every language your site serves
A bot that passes the first and fails the rest has an intent, not a disclosure.
Why this is the interesting one
Most Article 50 coverage stops at "put a notice on the chat". This requirement is in the same paragraph as the notice guidance, it is stated in mandatory terms, and in our experience almost nothing ships with it. If you build chat widgets for clients, it is also the most defensible thing you can charge for: it takes real understanding of the system, not a copy-paste.
It is check four of the four we put at the end of every recipe, and it is the one that fails.