My chatbot only answers fixed FAQs — do I still need an AI notice?

Short answer

It depends on what is actually working inside the widget. A rigid decision tree that only replays stored answers is not an AI system under the regulation; the moment a model interprets the question or writes the answer, it is one. From the outside that difference is frequently invisible — including to us.

Why that is

Article 50(1) presupposes an AI system interacting directly with people, so the trigger sits behind the chat window, not in it. Our check P1 is built around exactly that. For every provider in our detection data we record whether the product is an AI dialogue system by its nature — in which case we state the duty as applying — or a live-chat platform that can be staffed by humans just as well as by an AI bot. For the second group the scanner deliberately reports "needs review" instead of "not met": which mode you run is in your configuration, not in your page source. That restraint is the point. A guess in either direction would be the kind of false finding that costs more trust than it wins.

This is measured in check point P1 — Chatbot or AI dialogue system (Art. 50 Abs. 1).

What this means in practice

Open your chat provider's admin area and look at which feature is switched on. The useful question is not "is this AI?" but: does the widget answer free-form questions, does it compose answers itself, or does it only pick from a fixed list? And does it behave differently outside office hours? An AI-generated first reply at night is enough to trigger paragraph 1, even when humans answer during the day.

Check it on your own page

The check tells you which provider is embedded on your page, which of the two groups our data puts it in, and whether an AI disclosure was visible once the chat window was opened. Whether the AI feature is enabled on your account is something it cannot see — so it does not claim to.

Related questions

• The chat widget comes from a vendor — is it not their responsibility?
• How should an AI disclosure in a chatbot be worded?
• Where in the chat window does the AI notice have to appear?