What the AI Act asks of your chatbot

Did you build it, or buy it?

Built means you wrote it, had it built for you, or changed how a bought one works, for example by training it on new data. Giving it your documents to search does not count as changing it. Bought means off the shelf and used as supplied, whatever name is on it.

Does it carry your own name or trademark?

Do people see it as yours, under your name or brand, or as somebody else's? A tool you built only for use inside your own business counts as yours. A bought one your staff use under the supplier's name does not. The law counts as the maker (the provider) whoever builds an AI tool, or has one built, and then sells it or puts it into use under their own name or trademark. In the law's own words (Article 3(3)), that is whoever "develops an AI system ... or that has an AI system ... developed and places it on the market or puts the AI system into service under its own name or trademark". Both parts matter. A bought tool used unchanged under your own name meets only the second, and the law is not clear on whether that makes you its maker.

Does it hold a conversation with a person?

A website chatbot, a phone assistant, a support agent answering in your inbox.

Does it produce text, images, audio or video?

Drafting posts, writing replies, making images for ads, reading a script aloud.

Does it work out how somebody feels from their face, voice, body or the way they type, or sort people by how they look or sound?

For example scoring a caller's mood from their voice, or a camera reading facial expressions. Reading the sentiment of written messages or call transcripts is not this.

Was it already running on 2 August 2026?

Tools that create content and were already on the market before that date have until 2 December 2026 to mark what they create. If you bought it, what counts is when your supplier put it on the market, which can be well before you switched it on.