Ask around and you’ll hear that there’s now an official “EU icon” you have to stick on anything made with AI. That’s not quite right. There are three icons, not one. They come from a code that nobody is forced to sign. And the rules behind them care far more about whether people notice your label than about what it looks like.
So let’s sort out what the icons are, who they’re meant for, and whether a normal website should bother.
A tool, not legal advice. This guide explains public documents in plain language. If labeling is a serious matter for your business, check with a lawyer.
Where they came from
Article 50 of the AI Act splits the transparency work between two groups. The companies that build AI tools (providers) have to mark their output so software can detect it. That’s Article 50(2). The people who publish with those tools (deployers) have to tell readers and viewers about deepfakes and about AI-written text on matters of public interest. That’s Article 50(4).
The law says the disclosure must be clear and must come no later than the first time someone sees the content. What it doesn’t say is what a label should look like. A caption? A badge? A sticker in the corner?
The European Commission tried to answer that with a Code of Practice on marking and labelling AI-generated content, published on 10 June 2026. One half of it is for providers, the other for deployers. It’s voluntary. By the end of July, roughly 190 organizations had signed, according to the law firms that track it.
The icons sit in the deployer half.
The three icons
There’s one for content that is fully AI-generated. There’s one for AI-modified content, meaning a real photo, video or text that someone changed with AI. And there’s a basic icon that opens a second layer when you tap or hover over it, with more detail on what exactly AI did.
Audio is the odd one out, since nobody can see an icon in a podcast. For that case the Code suggests a short spoken disclaimer instead.
One finding from the testing stage is worth knowing. People understood the icons better when a word sat next to them, something like “modified”. A symbol on its own was harder to read. That’s a good hint even if you never touch the official icons.
Are you required to use them?
No.
The Code is voluntary, and even signatories aren’t forced to use the EU icons. They can use their own icon or label instead, as long as it follows the same design and placement principles.
What the law does require, for deepfakes and for unreviewed public-interest text, is a disclosure people actually notice. An icon is one way to get there. A clear sentence is another. Most small sites haven’t signed the Code and never will, and that’s fine.
Placement matters more than the picture
The Code’s placement principles apply to any label, official or not. Boiled down:
- People should get it without clicking, hovering or scrolling. The second layer of the basic icon is extra detail, not the main message.
- On video, it has to stay on screen long enough to read. A label that flashes for half a second doesn’t help anyone.
- Where possible, it should be part of the content itself: on the image, beside the video, at the top of the article.
- It has to be there at first exposure. If the image is the first thing a visitor sees, the label is too.
Here’s a quick way to check. Take a screenshot of just the image or video. Is the label in it? If not, it’s probably too far away.
Icons are for people, watermarks are for machines
These two get mixed up all the time.
Under the provider half of the Code, AI companies are expected to mark their output with at least two machine-readable layers, such as signed metadata plus an invisible watermark. Free-form text can get by with one. Detection is supposed to work across different tools by 2 February 2027.
That’s not your job if you run a website. You don’t need to watermark the images you publish. But there’s a side effect you’ll like. More and more AI files now arrive with markers already inside them, so it’s getting easier to spot what needs a visible label.
What actually needs a label
The icons don’t change the list of things that need labeling. Under Article 50(4), that list is short:
- Deepfakes. AI images, video or audio that look like real people, places, objects or events and could pass as genuine. Our guide Deepfake or not: where the line is has examples.
- AI text on matters of public interest, unless a competent person reviewed it and someone takes editorial responsibility. More in Do you need to label AI-written blog posts?
A cartoon mascot doesn’t need a label. Neither does an abstract banner or, usually, a product description. Plenty of sites label them anyway, because it costs nothing and readers appreciate it. That’s your choice, though, not a legal duty.
If you do use the icons on WordPress
Download the official files from the Commission’s Code of Practice page and use them as they are. Don’t redraw or recolor them.
Put a few words next to the icon, like “AI-generated image” or “Edited with AI”. Plenty of visitors won’t know the icon yet.
Be careful not to imply you’ve signed the Code if you haven’t. Using the icon doesn’t make you a signatory, so skip anything like “EU certified” or “EU approved”.
Then check your site on a phone. An icon that looks fine on a large desktop image can shrink to a dot on mobile.
How AuthorLabel helps
AuthorLabel shows a plain text label on the content you mark. AI images, video and audio get a small caption, such as “AI-generated image” or “Image edited with AI”. AI-written posts get a short note above the text. This is the kind of equivalent label the Code allows, and it sits right on the content.
It also helps with the harder part, which is knowing what to label. When you upload a file, the plugin reads the C2PA, IPTC and EXIF/XMP markers that AI tools leave behind and suggests a label. You confirm it, and the change is saved in the activity log.
You can rewrite the label text in Labels & notices. You can also link every label to your “How we use AI” page, so anyone curious can find out more with one click. That works a lot like the second layer of the basic EU icon.
In short
The EU icons come from a voluntary Code published on 10 June 2026. There are three of them: fully AI-generated, AI-modified, and a basic icon with a second layer. You don’t have to use them, and a clear label of your own is allowed. What counts is that people see it right away and right next to the content. Watermarks belong to the AI companies, not to you.
Sources
- Regulation (EU) 2024/1689 (AI Act), Article 50(2), (4) and (5).
- European Commission, Code of Practice on marking and labelling of AI-generated content (June 2026). digital-strategy.ec.europa.eu
- Freshfields, “EU AI Act unpacked #33: the final Code of Practice on transparency of AI-generated content” (June 2026). freshfields.com
- Paul, Weiss, “EU finalises transparency rules for AI-generated content” (August 2026). paulweiss.com
Last reviewed: November 2026. AuthorLabel is a tool, not legal advice.