AI Labelling Obligation for Your Website: What the EU AI Act Requires

The AI labelling obligation under the EU AI Act has applied since 2 August 2026, including to websites. But what does that actually mean?

The key facts at a glance

• The transparency obligation from Article 50 of the EU AI Act (Regulation 2024/1689) applies from 2 August 2026.
• It applies regardless of the channel, so a website is treated like any other. What matters is not where the content is published, but what is published.
• The most important case for websites: an AI chatbot has to be clearly recognisable as AI.
• You must label convincingly real AI images and deepfakes. You don’t have to label reviewed texts or obvious illustrations.

What is the AI labelling obligation?

The rule sits in Article 50 of the EU AI Act, the European AI Regulation (Regulation 2024/1689). Its transparency obligations apply from 2 August 2026.

The idea behind it is simple: people should not be deceived. Nobody should believe they’re chatting with a human when an AI answers. And nobody should mistake an AI image for a real photo. So it’s not about “AI is evil, label it everywhere”. It’s about avoiding deception in a few concrete situations.

One thing to take the pressure off: in some cases the EU AI Act also requires a machine-readable marking, like a watermark or metadata. But that part is not on you, it’s on the provider of the AI tool. For you, at most the visible marking applies, the kind a human can recognise.

Does this only apply to social media?

No, and that’s the most common misunderstanding. The AI Act is channel-independent. It depends on the type of content, not the place. An AI deepfake has to be labelled on Instagram just as much as when it’s embedded on your website.

Two things often get mixed up here:

• Instagram, TikTok, YouTube and others each have their own AI labels, but those are platform rules, not a legal obligation. When people see such a label, they quickly think “that’s the AI Act”. It isn’t.
• For example, on YouTube and Instagram, AI-generated content has to be flagged at upload, and the platform then shows a label on the content automatically. Every platform does this differently, by its own rules. So it’s worth checking once what each platform asks of you on top of the AI Act.
• The wave of warning letters over music licences on social media has nothing to do with the AI Act. That’s copyright. Please don’t lump them together.

Wasn’t the AI labelling obligation delayed?

Only in part, and not the part that affects the self-employed. The headlines recently said the AI Act had been delayed. But only the strict obligations for high-risk AI were postponed, into 2027 and 2028.

The transparency obligation from Article 50, meaning exactly the labelling of chatbots and AI content, stays on schedule for 2 August 2026.

What you have to label on your website, and what you don’t

Here are the situations that matter:

On your website …	Label it?
AI chatbot or support assistant	✅ Yes. Make it clear that an AI is answering, for example “You’re chatting with our AI assistant”. The most common case for the self-employed and micro-businesses. Rule of thumb: anything a user could mistake for a human has to be labelled.
Texts written with AI help (blog, product texts, about page)	❌ Mostly no. The text obligation mainly targets texts on public-interest topics like politics or society, for example news or investigative journalism. Ordinary marketing and product texts hardly fall under it, and as soon as a human reviews and takes responsibility for them, no label is needed anyway.
Convincingly real AI images or videos of people, places or events (deepfake style)	✅ Yes. Flag it as artificially generated. It doesn’t matter whether it recreates a real person or shows a fictional person who doesn’t exist at all. Both have to be labelled. For places this is still disputed. To be safe, label purely fictional places too.
Obvious AI graphics or illustrations (clearly not meant as real)	❌ No. No label needed. For example: comic-style drawings, clearly fantastical images.

Rule to remember: label the chatbot and anything that could look like it depicts reality, such as convincingly real images of people. Reviewed texts and obvious illustrations, stay relaxed.

One quick point on the chatbot: the obligation for it to identify itself as AI legally falls mainly on the provider of the tool. Many chatbot solutions therefore build the notice in already. Your job is then just to check that it’s actually visible. If you build your chatbot yourself or offer it under your own name, you count as the provider and are directly responsible.

A word on texts, because that’s where the confusion is greatest: labelling AI texts mainly targets content that informs the public on socially relevant topics, such as politics or the economy. Your product description, your about page or a guide in your field usually don’t fall under it. And even where the obligation could apply, it falls away as soon as a human reviews the text and takes editorial responsibility.

What you can safely ignore

• An “AI-generated” badge on every page? Not needed, and it undermines trust more than it helps.
• Marking every AI-assisted sentence individually? No. That isn’t necessary.
• Panic posts trying to convince you there’s a blanket labelling obligation for everything. The obligation is much narrower.
• Instead, a small note like “generated with AI” next to your deepfake image is enough, for example.

And an important bit of context: the strictest obligations mainly hit the providers of the AI systems, the tool makers. If you use ChatGPT to draft your newsletter, you’re in a far more relaxed position than the fuss suggests.

And what if you work with Jimdo Companion?

Companion is Jimdo’s AI that helps you build your website and write your texts. Legally, that counts as AI support, not as AI that could deceive your visitors. You review the results, publish them and stand behind the content. So the same logic as above applies: reviewed content needs no AI label.

Labelling only becomes relevant when an AI talks directly to your visitors, for example as a chatbot. Since Companion works in the background and doesn’t chat with your customers itself, no labelling obligation arises for you towards your visitors.

And what about images or graphics you create with Companion? Here too, the content counts, not the tool. A decorative graphic, a background or an obvious illustration needs no label. Only when an AI image convincingly shows a person, a place or an event does labelling kick in, exactly as in the table above.

The technical, machine-readable marking of such AI content is the job of the AI tool’s provider, not yours. Your job stays the visible notice wherever an image looks real and could deceive people.

The “how we use AI” page: nice to have

A short, honest page along the lines of “how we use AI” builds trust and fits a transparent business well. But it doesn’t replace concrete labelling where there really is a deepfake or an unreviewed informational text in play.

Your AI labelling checklist to tick off

Go through your website step by step. For each point: check quickly, tick the box, move on.

1. Chatbots and assistants

• Do you have an AI chatbot or support assistant on the site?
• If yes: is it clearly recognisable at first contact that an AI is answering?
• Notice built in? Template to copy: “Hi! You’re chatting with our AI assistant. For personal matters we’re happy to pass you to a human.”
• Is the notice clearly visible and accessible, so readable for screen readers and with good contrast?

2. Images and videos

• Do you use AI images or videos that look real (realistic people, places, events)?
• If yes: labelled as AI-generated, or replaced with a real or obviously artificial image?
• Obvious AI illustrations (graphics, cartoons, clearly not real)? Then nothing to do.

3. Texts

• Do you report on matters of public interest? Then the labelling obligation applies.
• Have your AI texts in the blog, shop or on subpages been reviewed editorially?
• Do you stand behind what’s written? Then you don’t need an AI label.

4. Trust (optional, but strong)

• Set up a short “how we use AI” note or page? Builds trust, but doesn’t replace concrete labelling.

5. Keep it clean

• Process noted down: who reviews AI content before publishing? For teams, a fixed rule saves discussions later.
• When in doubt on customer-relevant topics: a quick check with a lawyer.

Done? If points 1 and 2 are sorted, you’ve handled the essentials.

Getting your website ready? With Jimdo you build the foundation, legally sound basics, a clear structure, everything in one place, without wrestling with technical stuff. Start for free

Frequently asked questions

Does the AI labelling obligation only apply to social media?

No. It’s channel-independent and applies to a website just the same. It depends on the type of content, not the platform. On top of that, social media and other platforms may apply their own AI-labelling rules.

What do I have to keep in mind with an AI chatbot?

Your visitors have to be able to tell that they’re writing with an AI. A clear notice is enough.

When does the rule apply?

The transparency obligations from Article 50 of the EU AI Act apply from 2 August 2026.

Do I always have to label AI images?

Only if they look convincingly real, for example realistic people, places or events. Obvious illustrations need no notice.

Was the AI labelling obligation delayed?

No. Only the obligations for high-risk AI were postponed, to 2027 and 2028. The transparency obligation for chatbots and AI content applies as planned from 2 August 2026.

Do I have to label it if I build my website with Jimdo Companion?

No. Companion helps you build and write, and you review and publish the results. That counts as AI support and needs no AI label. Labelling only becomes necessary when an AI communicates directly with your visitors, for example as a chatbot, or when you use images that could look convincingly real.

What is a deepfake?

A deepfake is an AI-generated or AI-altered image, video or audio that shows real or convincingly real people, places or events and gives the impression of being genuine. The deciding factor is deception: if someone could take the result for a real recording, it’s a deepfake in the eyes of the law and has to be labelled as artificially generated. Obviously fictional or artistic depictions like comics or fantasy images don’t count.

Please note

Last updated: July 2026. Not legal advice: this article explains the rules in general, understandable terms, but does not replace legal advice. For anything that affects many customers, a quick check with a specialist lawyer is worth it.