Just a few years ago, the phenomenon known as "AI slop" was virtually unknown. Today, the internet is grappling with a massive influx of low-quality AI-generated content. A report by Kapwing, published in early June 2026, found that AI-generated content accounted for 60% of the content presented to new users and children in TikTok's "For You" section. Spotify is grappling with a similar problem, removing 75 million tracks deemed to be low-quality AI-generated content.
The growing scale of this phenomenon is addressed by, among other things, the transparency obligations set out in Article 50 of the AI Act, which will come into effect on August 2, 2026. This provision governs the obligations of providers and entities using certain AI systems. In the case of visible content labeling, the obligations will primarily rest with entities using AI systems.
Article 3(4) defines "users" as "(...) a natural or legal person, public authority, agency or other entity that uses an AI system over which it has control, except where the AI system is used in the course of a personal, non-professional activity." In the context of the marketing industry, this obligation will apply to all entities that use AI in the creation of graphic, audio and video content, for professional or commercial purposes.
Who do the obligations apply to?
In the marketing industry, it is crucial to determine who is using an AI system—that is, who is using it under their control as part of their professional or commercial activity. Such an entity may include, in particular, an advertiser, marketing agency, online store, publisher, influencer, or freelancer. If an employee uses the system under the direction and control of a company, the entity using it is generally the company, not the employee. The involvement of an agency or external contractor does not automatically transfer this status. The decisive factor is whose benefit, responsibility, and control the AI system is being used for. If an agency uses it independently and under its own control, it may also be considered a user.
What content needs to be marked?
The obligation under Article 50(4) primarily covers deepfakes, i.e., images, audio recordings, or videos generated or manipulated by AI that resemble real people, objects, places, entities, or events that may be mistaken for authenticity. In marketing, this could include, for example, a photorealistic image of a famous person endorsing a product, a realistic visualization of actual packaging disguised as a photograph, or a synthetic recording of the voice of a specific famous person.
Assessment is not limited to whether the advertisement accurately portrays the product's features. It's also crucial to consider whether the viewer could recognize the synthetic material as a real photograph or recording. Therefore, even computer-generated graphics that faithfully represent the product being offered may require labeling if they appear to be an authentic photo. The context of the publication and the target audience must be considered; messages aimed at children, seniors, and those with lower digital literacy require particular caution.
When will marking not be needed?
As a rule, standard editing activities, such as adjusting brightness, sharpness, color, or sound, do not trigger the obligation, provided they do not fundamentally alter the content or its meaning. Materials that are clearly fictional, cartoonish, or caricature-like, and that the recipient would not recognize as a representation of reality, may also fall outside the definition of deepfake. However, each case requires an assessment of the entire message – a humorous or fantastical element alone does not automatically exclude the obligation.
How to properly mark material?
The information should be clear, visible, and accessible no later than the first time the recipient sees the material. A statement such as "Content generated by AI" or "This material has been modified using AI". The labeling should be adapted to the format: in graphics, it should be legible and contrasting, and in audio or video material, it may also require an audio message. It is important to consider alternative text and the needs of people with disabilities. Technical metadata provided by the system provider does not replace the visible disclosure required from the entity using the material. The AI label also does not replace the information that the material is an advertisement or sponsored content.
What should the marketing department do?
Before publication, it is worth implementing the following procedure:
- determine whether and to what extent AI has been used; assess whether material resembles a real person, product, place or event;
- check the possibility of misleading the recipient;
- decide on the marking method and keep proof of verification.
Contracts with agencies and graphic designers should include obligations to disclose AI use, maintain technical markings, properly describe content, and provide compensation for damages in the event of infringement. Contract provisions regulate liability between the parties, but do not automatically change who is the entity using AI under the AI Act.
Material generated before 2 August 2026 is not subject to mandatory retroactive marking, although the Commission encourages voluntary marking where possible.
This article is for informational purposes only and does not constitute legal advice.
Legal status as of July 30, 2026.
author:
series editor:
