The use of artificial intelligence in e-commerce and marketing is becoming standard practice. However, the way AI-generated or AI-modified content is used may give rise to specific legal obligations. This applies in particular to product images, marketing materials and other content that may be perceived as an authentic representation of reality.
The AI Act introduces new obligations concerning the labelling of content generated or modified by artificial intelligence. This does not mean, however, that every graphic created using ChatGPT, Midjourney or another AI tool must be labelled as “AI-generated”. The key factors are the nature of the content, how it is used and the risk that recipients may perceive it as authentic.
As of 2 August 2026, the provisions of Article 50 of the AI Act concerning transparency obligations for certain AI systems apply. The European Commission has also published guidelines explaining when businesses are required to inform recipients that AI has been used.
The AI Act and the Labelling of AI-Generated Images
Article 50 of the AI Act introduces transparency obligations both for providers of AI systems and for deployers of AI systems. A business that uses an AI system in the course of its activities may qualify as a deployer in this respect, including where it operates a website and publishes AI-generated or AI-manipulated content on that website.
Under the AI Act, a “deep fake” means AI-generated or AI-manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a recipient to be authentic or truthful.
Therefore, the mere use of artificial intelligence does not, in itself, trigger an obligation to label content. The decisive question is whether the recipient may be misled as to the authenticity of the image presented.
The European Commission confirms that the assessment should take into account, among other things, the degree of similarity, the content and message conveyed, the intended or reasonably foreseeable use of the material, as well as the context and expectations of recipients.
Does Every AI-Generated Graphic Have to Be Labelled?
No. The AI Act does not establish a general obligation to label every graphic, illustration or photograph generated using artificial intelligence. In particular, not every AI-generated marketing graphic will constitute a deep fake.
For example, a business may use an AI-generated illustration: on its homepage; as an illustration for an article; as a background for a marketing campaign; to create or reinforce a particular brand identity.
Where the context clearly indicates that the material is a creative work, illustration or marketing content, and the graphic does not realistically depict a specific existing person, product, place or event in a manner that could make it appear to be an authentic photograph, the mere fact that it was generated using AI does not necessarily trigger a labelling obligation under Article 50(4) of the AI Act.
It should not, however, be assumed that describing something as “marketing material” automatically excludes the obligation. If an advertisement uses an AI-generated image of a real person, actual event or existing object in a way that may be perceived as an authentic representation of reality, a labelling obligation may arise.
AI-Generated Images vs. AI-Modified Images
A distinction should be made between two situations:
AI used for standard image editing;
AI used to create an image from scratch.
The European Commission’s guidelines indicate that there are situations in which AI performs only an auxiliary function in standard editing and does not result in a substantial alteration of the input data or its semantics.
If the product remains the same product and the modifications do not materially change the meaning or character of the photograph, the image may fall within the category of standard editing.
For example, a business may have an authentic photograph of a garment and use AI to: remove the background; adjust colours; improve image quality; enhance the overall aesthetics of the image. In such a case, it should not automatically be assumed that the photograph becomes a deep fake simply because an AI tool was used to edit it.
The situation is different where no authentic photograph exists and the entire image is generated by AI. If the resulting graphic is then used as a photograph of a specific product and may appear to be an authentic photograph of that product, there is a significantly greater risk that it will qualify as a deep fake.
An AI-Generated Image of a Real Product – A Particular Risk for Online Stores
The greatest uncertainty may arise in relation to product images used by online stores.
Suppose a business sells clothing. The product actually exists and is part of the business’s offering. However, instead of taking a traditional photograph of the product, the business uses an AI generator to create a photorealistic image of the garment.
The generated image may make the product appear as though it had been photographed in a professional studio. This situation differs from using AI to create an abstract marketing graphic. The image is displayed in connection with a specific product, and its primary purpose is to visually present that product to consumers. The recipient may therefore reasonably assume that they are viewing an actual photograph of the product. Moreover, the fact that the product actually exists does not automatically eliminate the potential labelling obligation. In certain circumstances, it may even strengthen the case for labelling.
If a business displays a photorealistic AI-generated image next to the name of a product, a consumer may reasonably understand it to be an actual photograph of a specific product.
How Should an AI-Generated Image Be Labelled?
If the material qualifies as a deep fake, the business should disclose this fact in a manner that is clear, noticeable and understandable to the recipient.
The European Commission indicates that the label should be visible to the recipient without requiring the use of special technical tools. Merely embedding a technical marker that is invisible to the user is not sufficient to fulfil the obligation to disclose a deep fake.
In practice, businesses should ensure that the information is displayed directly next to the image or in another location where the consumer receives the information before, or at the latest upon, their first interaction with the content.
At our law firm, we advise businesses on the legal aspects of using artificial intelligence, including obligations arising under Article 50 of the AI Act. We analyse, among other things, AI-generated photographs and graphics, marketing materials and the use of AI in e-commerce.
We can conduct an audit of the materials used on your website, assess which materials require labelling, and develop practical rules for classifying and labelling AI-generated or AI-modified content.
This article is for informational purposes only and does not constitute legal advice.