The answer in thirty seconds
This is the only AI Act obligation that already applies to almost every business: since 2 August 2026, you have to say when an AI is talking to a person, and flag certain content it produces. It is not complicated, it is not expensive, and it is what will be looked at first.
The three cases where you have to say so
A system that interacts directly with a person must be designed so that they know they are dealing with an AI — unless this is obvious to someone reasonably observant. That means the website chatbot, the telephone assistant, the smart answering service.
Synthetic content — audio, image, video, text — must carry a machine-readable marking, so that it is detectable as artificial. This obligation falls on the provider of the generating system, not on you as its user.
Deepfakes and certain published texts. If you put out an image, a sound or a video that imitates a real person, place or event, you must state that it is artificial. And if you publish an AI-generated text to inform the public on a matter of public interest, you have to disclose it — unless that text has undergone human editorial review with an identified person responsible for it.
What this means for your website
Three checks, ten minutes:
- Does your chatbot announce, before the first exchange, that it is automated? One sentence is enough: ‘You are chatting with an automated assistant. To speak to a person, write to…’. Visible, not buried in the terms and conditions.
- Are your AI-generated visuals flagged when they show people or places that could be taken for real?
- Do your AI-written blog posts on matters of public interest carry a notice, or the name of a human reviewer?
What does not apply — and what you will be told it does
There is no retroactivity: content published before 2 August 2026 does not have to be relabelled.
A text reviewed and owned by an identified person does not have to be flagged as AI-generated. Human editorial responsibility is expressly provided for in the text.
A notice is not required when the use of AI is obvious: nobody needs to be told that a photo filter is automated.
The date coming up
On 2 December 2026, generative systems placed on the market before 2 August 2026 must in their turn comply on machine-readable marking. If you provide such a system, that is your deadline. If you use one, this is the right moment to ask your vendor how far along it is.
A simple route to compliance
The Commission has approved a Code of Practice on the transparency of AI-generated content, the final version of which is dated 10 June 2026. It is voluntary: signing it offers a simple route to demonstrating compliance, not signing it means having to prove equivalent means. Around 190 organisations had signed it by the end of July 2026 — 82 as providers, 152 as deployers — and signature remains open.
Common mistakes
‘That is for our supplier to handle.’ The technical marking, yes. Informing people on your website, no: it is your page, your visitor, your responsibility.
‘We will put the notice in the terms and conditions.’ The information has to be given at the time of the interaction, in a clear and distinguishable way.
Your next step
Compare this page with how you actually work, then ask the community if anything is still unclear.
Ask the community →Reference sources
Going further
This page is based on the consolidated text of Regulation (EU) 2024/1689 in its version of 27 July 2026. Verified 26 August 2026. General information: it does not constitute individual legal advice.