Almost every business with a chat widget has one question to answer, and it's simpler than it sounds.
If visitors to your site can chat with something that isn't a person, whether that's a basic FAQ bot or a more capable assistant built on ChatGPT, the EU AI Act has one clear expectation: people need to know they're talking to AI, not a human, unless it's obvious from the context.
That's the whole obligation for most chatbots. It's a transparency requirement, not a high-risk one. A line in the chat window, something like "You're chatting with our AI assistant", usually covers it. Many chat tools already do this by default. Worth actually checking yours does, rather than assuming.
Where it moves into different territory: if your chatbot is making decisions that affect someone, approving or declining a request, screening a job candidate, deciding who gets an offer and who doesn't, that's no longer just a chat interface. It's an automated decision system, and the rules that apply depend on what it's deciding.
A support or sales chatbot that answers questions and hands off to a human for anything that matters: transparency label, done. A chatbot making real decisions about a real person: worth running through the checker properly, since the answer depends on exactly what it's deciding.
Not sure where your business stands?
Answer a few questions about the AI tools you use and find out in about two minutes. Free, no sign up needed.
Check your risk level