AI-generated visuals have become part of everyday marketing, social media and e-commerce. Businesses are using them to create advertising assets, product backgrounds, people, interiors and even entire scenes that never existed in real life. This raises an increasingly important question: do these images need to be labelled?
Since 2 August 2026, transparency requirements under Article 50 of the EU AI Act have applied to certain types of AI-generated and manipulated content. The aim is to make it clear when content that appears authentic has actually been created or altered using AI.
The most important point is that not every image created or edited with AI needs a visible label.
When does an AI-generated image need to be labelled?
For businesses, one of the most relevant parts of the AI Act concerns so-called deepfake content.
Under the regulation, a deepfake is AI-generated or manipulated image, audio or video content that resembles an existing person, object, place, entity or event and could falsely appear to be authentic or truthful.
In practice, this could include:
- a realistic AI-generated person who could be mistaken for a real person;
- an AI-generated face swap involving a real person;
- a real photograph of a location that has been substantially altered to show something that was never actually there;
- a realistic-looking event or situation that never took place.
The European Commission, for example, gives the example of a genuine photograph of an empty apartment that is furnished using AI in a way that makes the final image look like an authentic photograph.
In these situations, a business using the content needs to consider whether it should clearly disclose that the image was generated or manipulated using AI. Article 50 of the AI Act requires those deploying deepfake content to disclose its artificial nature.
Does every AI-edited image need to be labelled?
In short: no.
The fact that AI was used somewhere in the editing process does not automatically mean the image falls under the deepfake disclosure requirement.
European Commission guidance indicates that standard editing assistance may fall outside the scope of the transparency requirements for AI-generated or manipulated content.
So if AI is used only for minor retouching, colour correction, noise reduction or a similar technical adjustment, and it does not materially change what the image depicts, it is not treated in the same way as realistic synthetic or substantially manipulated content.
The more useful question is not “Was AI used?”, but rather “Did AI change the content in a way that could make someone believe they are seeing something authentic?”
What about creative or obviously fictional images?
The AI Act also makes specific allowances for artistic, creative, satirical, fictional and similar works.
Where deepfake content forms part of an obviously creative or fictional work, the transparency requirement still applies, but the disclosure can be presented in a way that does not interfere with the enjoyment or presentation of the work.
This means a realistic advertising image and an obviously fantastical AI illustration will not necessarily be treated in exactly the same way.
Who is responsible for labelling AI-generated images?
There are two separate responsibilities to understand.
First, providers of AI tools are responsible for technical measures that make it possible to identify content as AI-generated or AI-manipulated.
However, if a business publishes that content and it falls within the definition of a deepfake, technical marking alone is not enough. The audience must also be clearly informed that the content has been generated or manipulated using AI.
Put simply: the AI provider is responsible for technical marking, while the business publishing the content is responsible for clearly informing the viewer when disclosure is required.
How can AI-generated images be labelled in practice?
The EU has already published official AI content labels that businesses can use free of charge. Their use is optional, but they can provide a practical way to make it clear that an image has been generated or altered using AI.
However, adding an icon does not automatically mean a business is fully compliant with every requirement under the AI Act.
If you use AI-generated visuals regularly, you also do not need to create every label manually.
Shopify stores can use the EU AI Icons & Labels app, which allows merchants to label AI-generated or edited product images and control how those labels are displayed on the site.
WordPress websites can use the AI Image & Video Disclosure Labels plugin, which lets site owners choose which images or videos should display an AI disclosure label.
These tools make the technical side of labelling easier, but they do not decide whether a particular piece of content is legally required to carry a disclosure. The business still needs to assess how the image was created or altered and how it is presented to the audience.
What should businesses do now?
If you already use AI in marketing, product visuals or social media content, it is worth carrying out a simple review of where and how it is being used.
Pay particular attention to images where AI has:
- created a realistic-looking person;
- altered the appearance of a real person;
- materially changed a product, location or environment;
- created a realistic situation that never actually happened.
For this kind of content, it is useful to establish a simple internal process in advance: who decides whether disclosure is needed, and how that disclosure will be shown on your website, in advertising or on social media.
The key question is not whether AI was used, but what it changed
The new AI Act rules do not mean that every image touched by AI needs a label.
For businesses, the important distinction is between routine AI-assisted editing and realistic AI-generated or materially altered content that could be mistaken for something authentic.
If AI is already part of your regular marketing workflow, now is a good time to establish a clear internal process: which content needs to be reviewed, who makes the decision on disclosure, and how that disclosure will be displayed.
This article is not legal advice. Where the situation is unclear or potentially higher risk, it is worth seeking advice on the specific use case.
Frequently asked questions
Do all images created with ChatGPT, Midjourney or other AI tools need to be labelled?
No. The AI Act does not require every AI-generated image to carry a visible label. Businesses should assess whether the content falls within the definition of a deepfake and whether it could be mistaken for authentic content.
Does an AI-retouched product image need to be labelled?
Not necessarily. If AI was used only for standard editing and did not materially change what the image depicts, that does not automatically make it deepfake content.
Can businesses use their own AI label?
Yes. Using the EU’s official icons is optional. What matters is that, where disclosure is required, it is clear and meets the relevant AI Act requirements.
Does an AI-labelling app automatically make a business compliant with the AI Act?
No. These tools can help display a label, but the business remains responsible for deciding whether a particular piece of content requires disclosure and whether that disclosure is appropriate.
