Skip to content
Get a quote
Get a quote

AI rules

Do you need to label AI images and videos in ads in the EU?

Since 2 August 2026, a business that shows people in the EU a realistic AI-generated image or video has to say it is AI. This applies to companies outside the EU too, as long as the ad reaches EU audiences. Not every graphic needs a label: drawings and obviously fantastical scenes are usually fine. Here is what the law, the European Commission guidelines and the platforms say, with small business examples.

Published 7 min read

In short

  • Article 50 of the EU AI Act applies from 2 August 2026: realistic AI images or videos that someone could take for real must be clearly labelled.
  • It applies to any business using AI professionally, including sole traders and companies outside the EU whose content reaches people in the EU.
  • Drawings, obviously fantastical scenes and routine edits (colour, sound clean-up) usually need no label.
  • A label does not excuse a misleading ad: the product in an AI image must not look better than it really is.

Short answer

Since 2 August 2026, Article 50 of the EU AI Act applies. For a business that publishes AI images and videos, paragraph 4 matters most: whoever shows people a so-called deepfake has to disclose that it was generated or changed by AI.

Under the AI Act, a deepfake is any realistic image, audio or video that resembles existing people, objects, places or events and that someone could wrongly take for real. That definition covers many ordinary ads.

What you publishLabel it?
Realistic product photo generated or changed by AIyes
AI photo of a dish for a menu or an adyes
Virtual staging of an empty flatyes
Video with realistic AI people, fictional ones includedyes
Real product on an AI-generated backgroundusually not, if the product is shown honestly
Drawing, animation, obviously fantastical sceneusually not
Colour correction, background removal, sound clean-upno

This is a practical summary, not legal advice. For a large campaign or a borderline case, ask a lawyer.

Who has to do it

The AI Act splits the duties into two groups. Tool providers (such as Google or OpenAI) have to mark what their tools generate in a machine-readable way, for example in the file’s metadata. Deployers, meaning any business that uses AI in its work, have to disclose deepfakes to the people who see them. Only private, non-professional use is exempt.

So a sole trader who posts an AI-generated cake photo on Instagram for customers in the EU falls into the second group.

It does not matter where your business is based. The rules can reach advertisers outside the EU whose AI-generated content is shown to people in the EU.

The EU changed parts of the AI Act in 2026 with the so-called Digital Omnibus package. It did not postpone the duty to disclose deepfakes. It only gave tool providers a few extra months for the technical marking. If you read that “AI labelling was postponed”, that is about the tool makers, not about your ads.

When an ad is a “deepfake”: examples from the guidelines

In July 2026 the European Commission published guidelines on Article 50. They are not binding, but regulators are expected to follow them. According to law firm summaries (for example Davis+Gilbert), they include these advertising examples:

Deepfake, label it:

  • an AI product image that makes the product look different, more appealing or higher quality than it really is
  • an AI-generated celebrity recommending a product
  • a synthetic influencer testing a real product.

Usually not a deepfake:

  • talking mice arguing about cheese in a cheese ad
  • a real car on an AI-generated background, if the ad does not mislead about the car
  • audio adjustments that do not change the words or the way someone speaks.

The guidelines also contain an important point for advertisers. The exception for “evidently artistic or fictional” works reduces the duty, but in mixed content the commercial character usually prevails, so ads rarely qualify. A fictional person in an ad is still a deepfake if they look realistic.

How to label it properly

The label has to be clear and visible no later than the first time someone sees the content. Metadata in the file or a line in your terms and conditions is not enough. In practice:

  • photo: a caption or a small label on the image, for example “AI-generated” or “AI visualisation”
  • video: a caption in the frame at the start or throughout, plus a note in the description and the platform’s AI toggle
  • property listing: “virtual staging” next to the photo, ideally with a real photo of the empty room beside it.

The Commission has also prepared common EU icons for labelling AI content, described in a voluntary code of practice published in June 2026.

On my own website, every AI image in the “Before and after” section carries an “After · AI” label, and the sample videos say “Fictional brand · AI-generated”.

A label does not excuse a misleading ad

The AI Act is one set of rules. EU consumer law is another, and it applies on top. A photo that makes a product look better than it really is can be a misleading commercial practice under the Unfair Commercial Practices Directive, which every EU country enforces through its own consumer authority.

A “photos are for illustration only” disclaimer does not fix that either. In Poland, the consumer authority UOKiK challenged exactly this kind of clause in the terms of Deichmann’s online shoe shop, and a court upheld the decision. An “AI” label says how the image was made. It does not excuse showing the product dishonestly.

My rule for client images: AI can change the background, the light, the scene and the mood, but not the product. The mug keeps its glaze colour, the cake keeps its portion size, the flat keeps its windows and floor.

Platform rules

Platforms have their own rules on top of the AI Act, and they apply worldwide:

  • Meta (Facebook, Instagram): adds an “AI info” label to ads made with its own generative tools, and to ads where it detects AI markers in the file. For political and social issue ads you have to disclose AI yourself.
  • TikTok: realistic AI content has to be marked with the “AI-generated content” toggle. Unlabelled content can be removed.
  • YouTube: you have to disclose realistic synthetic content, such as a scene that never happened. Animation, filters and scripts written with AI need no disclosure, and the label does not reduce reach.
  • Google Ads: since July 2026 there is an AI label setting for ads. Google notes that the setting alone does not guarantee compliance with EU law.

Do not strip the markers that AI tools add to files (such as C2PA). They let platforms recognise the content and label it for you.

When the fines start

Breaking Article 50 can cost up to €15 million or 3% of worldwide annual turnover, and for small and medium-sized businesses the lower of the two applies. Enforcement is national: each EU country has its own authority. In Poland, for example, complaints and fines will be possible from 28 October 2026.

Checklist for a small business

  • Realistic AI photo or video for EU audiences? Add a visible label when you publish it.
  • Does the product look the way it really does? If not, change the image, not the caption.
  • Virtual staging of an interior? Label it and show a real photo too.
  • TikTok and YouTube? Switch on the AI content toggle.
  • Working with an agency? Agree in the contract who labels the material and who owns it.
  • Keep a note of which tool made each image and when. It helps if a client or a regulator asks.

Sources

Need AI photos or videos that follow the rules?

Everything I make with AI is labelled, and your product looks the way it really does. Tell me what your business does and I will make your first sample photo free of charge.

See AI photos and videos

FAQ

Frequently asked questions

My business is in the US or UK. Does the AI Act apply to me?

It can. The transparency rules look at where the content reaches people, not where it was made. If you run ads or post content aimed at customers in the EU, the labelling duties apply to that content. Content shown only outside the EU is not covered by the AI Act, but platform rules still apply everywhere.

Do I need to label text written with AI, such as a product description?

Usually not. The text rules cover AI text that informs the public on matters of public interest, such as health or politics, and even then text reviewed by a person who takes editorial responsibility is exempt. A product description written with AI and checked by you needs no label. It does need to be true.

Do I need to label a photo of my real product on an AI-generated background?

According to the Commission guidelines, usually not, as long as the ad does not mislead about the product itself. The guidelines use a real car on an AI background as an example. It is different when AI changes the product: its colour, portion size or material quality.

Who is responsible when an agency made the ad?

Lawyers disagree. Some say the duty sits with whoever decides to use AI, which is often the agency rather than the brand. The safest option is to agree in the contract who labels the material and to check that the label is visible when it is published.

Do I own an AI-generated image?

You can use it under the tool's terms and your contract with the supplier. But in the EU, an image generated entirely from a prompt is usually not protected by copyright, because a work has to come from human creativity. In practice you cannot stop a competitor from making a very similar image.

Next step

What shall we build for your business?

A website that turns visitors into clients, automation that gives you back hours, or AI that answers your clients. Tell me what you need.

Write to me contact@sitebest.eu

I usually reply within a few hours · individual quote