Do I have to label AI text on my website? What the AI Act says

Usually not: your services, prices and product pages are out of scope. When a blog post does need an AI label, and how you can avoid needing one.

By Kintsu editorial team · 4 November 2026

Usually not. Since 2 August 2026, the EU's transparency rules for AI have applied (Article 50 of the AI Act). They require a disclosure for chatbots, for deepfakes and for AI-generated text that informs the public on matters of public interest, such as health or new regulations. Even then, you don't need a label if someone genuinely reviewed the text and you or your company openly take responsibility for it. According to the European Commission, your service pages, prices and product pages are out of scope.

What exactly does the AI Act say?

The law has two roles. The provider develops an AI program or places it on the market under its own name: OpenAI for ChatGPT, or a company that builds its own chatbot. The user (in the law: the deployer) uses AI for work: you, when you write a page for your site with ChatGPT.

Article 50 sets out these main duties:

Chatbots (provider): people talking to an AI must be told so, unless it is obvious.
A hidden mark (provider): images, audio, video and text from an AI get a mark that computers recognise, such as a digital watermark. That is the job of OpenAI, Google and the like, not yours.
Deepfakes (user): anyone who uses AI to create realistic images, audio or video of existing people, places or things must say they were AI-generated or manipulated.
Text on matters of public interest (user): anyone who publishes such a text made with AI must disclose it, unless the exception applies (see below).

The disclosure must be immediately visible and clear, including for people with visual impairments.

Has anything changed in the meantime?

Not for you. In late 2025, the European Commission proposed simplifying the AI Act with a "digital omnibus". It became law in July 2026. It only gives makers of AI programs that were already on the market before August 2026 extra time, until 2 December 2026, to add the hidden mark. It also postpones the rules for high-risk AI, such as AI that screens job applicants (European Parliament). The rules for anyone publishing AI-generated text or images have applied since 2 August 2026 (European Commission).

Content published before 2 August 2026 doesn't need to be labelled retroactively. An older AI-generated text that you only publish now does fall under the rules.

Does my small business blog fall under it?

On 20 July 2026, the European Commission published an explanation of Article 50, the "guidelines" (European Commission). They are not binding, but show how regulators read the text. A text is in scope if it is:

published: anyone can read it. An email to a client is not;
meant to inform: it communicates knowledge, opinions or facts;
about a matter of public interest: something that concerns society as a whole and merits public debate. The Commission lists politics, public administration, justice, security, public health, the environment, consumer safety, and economic or scientific developments that are debated in public.

That sounds broad, but the Commission's examples make it more concrete. Out of scope: AI-generated text in advertising or product descriptions, as long as it makes no claims about, for instance, health, safety or sustainability. In scope: an article on a lifestyle website comparing the effect of diets on a disease.

For a small business, that means:

Your services, prices, about page and product pages: no label needed.
A blog about your trade ("how to look after an oak staircase"): usually not.
A blog about health, safety, the environment, tax or new regulations ("new insulation requirements", "what changes for renovation grants"): possibly. Then either label it, or make sure the exception really applies.

When do you not need a label?

Two things must both be true:

Someone genuinely reviewed the text. A person who knows the subject reads it critically and at least checks the facts. A spell check, an agreement that exists only on paper, an automated check or a quick click on "approve" is not enough.
Someone openly takes responsibility. You or your company carry final responsibility. Put the name and contact details of the person responsible on your site, for example in your legal information.

Note: if you have AI rewrite the text after the review, the exception no longer applies. Review it again.

What does such a notice look like?

If the exception applies, add something like this to your legal information:

Responsible for the content of this site: [your business name], represented by [name], managing director. [Address], [email], enterprise number [number].

Under the text itself, a line like this is not required, but it is honest and helpful for your readers:

Written with help from AI, reviewed by [name], [role].

If the exception does not apply, label the text right under the title, before the text begins:

This text was generated with AI.

There are also EU icons for this. They are free, open for anyone to use and optional. An icon on its own does not prove you comply (European Commission).

What about images or a chatbot on your site?

A deepfake is AI-generated or manipulated image, audio or video content that resembles existing people, objects, places or events and would falsely appear authentic. According to the Commission, an AI image that makes a product look better or different than it really is counts as one. Placing a real product against an AI background, adjusting colours or removing a passer-by usually does not.

And an artist's impression of a design, such as an AI image of what a kitchen could look like before it's built? That is usually not a deepfake, as long as it is clear that it is a design: viewers don't expect a real photo. Add "design visual, created with AI". If you show it among your finished projects, as if you really built that kitchen, you are misleading people. Apart from the AI Act, that can also be misleading advertising.

For a chatbot, the duty lies with the provider. If you use an off-the-shelf chatbot, the vendor normally takes care of the notice. If you build one yourself or put it into service under your own name, you may be the provider. Either way, check that the first message says so. The FPS Economy gives this example: "You are chatting with an AI assistant. Replies are generated automatically." (FPS Economy, in Dutch). According to the Commission, a general line somewhere on your site, such as "this website uses AI", is not enough.

What are the risks?

Fines can reach €15 million or 3% of worldwide annual turnover. For small and medium-sized businesses the lower of the two applies: with an annual turnover of €1 million, that is at most €30,000. Fines must also be proportionate to the breach. As of 4 November 2026, we could not find an official source stating which Belgian authority enforces these rules.

Checklist for your website

List where you use AI: texts, translations, images, a chatbot.
Service and product pages: no label needed, but do review them.
Blog posts on health, safety, the environment, money or regulations: have someone who knows the subject check the facts and put a name on the text. Or label it.
State in your legal information who is responsible for the content.
No more AI rewriting after the review.
AI design visuals: add "design visual, created with AI", and keep them apart from your real projects.
A chatbot: "AI assistant" in the very first message.

At Kintsu, AI writes your texts and nothing goes live until you've approved it. So do read your texts properly. For texts on regulations or health the law requires it; for everything else it's simply good practice (how AI works at Kintsu). Whether Google penalises AI texts is covered in Does Google penalise AI-written content?

Written with help from AI, reviewed by the Kintsu editorial team. This article is general information, not legal advice. If you are unsure about a specific text, ask a lawyer.

Sources

Accessed on 4 November 2026.

See it first, then decide

Would you rather see it than read about it?

We build your new site first, in three languages and with no setup fee. If you like it, you start a subscription.

Get your free preview