Labelling AI-generated images and content (Article 50)

What the marking duty actually covers

The deployer duty in Article 50 is about content that imitates reality. The concern is imagery, audio or video that a person could reasonably take to be a genuine recording of real people or real events, when in fact it was generated or manipulated by AI. In an online store, the clearest example is a synthetic human: an AI fashion model wearing your product, a face-swapped testimonial, or a virtual try-on that places a realistic person in your item.

The reason for the rule is straightforward. A shopper makes decisions based on what looks real. If a lifelike model or scene is actually synthetic, they deserve to know, because it changes how they read the image. The label does not have to be heavy handed. It just has to make the AI origin clear to someone looking at the image.

What falls outside the rule

Just as important is what you do not need to label, because over-labelling clutters your store and trains shoppers to ignore the notices that matter. The following are generally out of scope for the deployer marking duty:

• Plain AI product shots. An AI-assisted image of the product itself, with no synthetic person imitating reality, is not the target of the rule.
• Cleaned-up backgrounds. Removing or replacing a background, straightening, or relighting a genuine product photo is routine editing.
• Upscaling and retouching. Sharpening, denoising or enlarging a real photo does not turn it into content that imitates reality.
• Abstract or clearly stylised graphics. Illustrations and obviously artificial visuals are not going to be mistaken for a real photograph.

The test to keep in mind: could a reasonable shopper mistake this for a real photo or recording of real people or events? If yes, and AI made or altered it, label it. If no, you are almost certainly fine.

Where the machine-readable marker fits

You may have read that AI content must carry a machine-readable marker embedded in the file. That obligation sits with the provider of the AI tool that generated the content, not with the store using it. As the deployer, your duty is the perceptible one: a clear, human-visible label that a shopper can actually see. So do not get stuck trying to inject invisible metadata into your images. Focus on the visible label, which is the part that is yours to get right.

How to label it in practice

A good label is close to the content, readable, and plainly worded. Some practical guidance:

• Put it near the image. A caption or badge on or beside the affected image works far better than a blanket line in the footer that no one connects to a specific picture.
• Say it plainly. Wording such as AI-generated image or This model is AI-generated tells the shopper what they are looking at without jargon.
• Be consistent. Use the same label style wherever synthetic imagery appears, so shoppers learn to recognise it.
• Translate it. Show the label in the EU languages you sell in, so it is understood by the shopper who sees it.
• Keep it live. Check the label still renders after theme edits, so it does not quietly disappear.

A word on AI-written text

Text follows a narrower rule than images. The deployer marking duty for text is aimed at content published to inform the public on matters of public interest, such as news or health topics. An ordinary product description written with AI help and reviewed by a person is commercial content and generally needs no disclosure. If you run a blog that covers public-interest subjects with AI-written articles, that is where the text duty is more likely to come into play.

How Aclara handles it

Aclara adds an AI-content label to your product template or the relevant page as a theme block, translates it into 24 EU languages, and reads your live theme to confirm the label is really showing. Because it is built on the deployer view of Article 50, it flags the synthetic imagery that needs a label and does not push you to mark content that is out of scope. It uses a single read-only theme permission and stores no customer data.

Frequently asked questions

Do all my AI-edited photos need a label?

No. The marking duty targets AI content that imitates real people or scenes, such as synthetic models or virtual try-on. Plain AI product shots, tidied backgrounds and upscaling are generally out of scope.

What counts as AI content that needs disclosure?

Imagery a shopper could reasonably mistake for a genuine photograph of a real person or event, for example an AI fashion model wearing your product or an AI-manipulated scene of people.

Do I need a hidden machine-readable marker in the file?

That marker is a duty for the AI tool that generated the content, not for you as the store. Your side is a clear, human-visible label near the image.

What about AI product descriptions?

Ordinary commercial copy written with AI help and reviewed by a person generally needs no disclosure. The text marking duty is aimed at content on public-interest topics, not shop descriptions.