The AI labeling requirement has been in effect since August 2, 2026 and affects companies that publish AI content. The EU AI Act requires visible labeling for deepfakes and certain AI texts. Exceptions apply for other content. This article shows you which marketing content is affected and what you need to do now.

Many marketers have long been using AI for texts, images, and videos. At the same time, they want to remain visible in AI search. For this, you need clear rules for handling AI content. We'll explain where the requirement applies and where it doesn't.

What is the AI labeling requirement in the AI Act?

The AI labeling requirement is found in Article 50 of the EU AI Act. The article obligates providers and deployers of certain AI systems to greater transparency. People should recognize when they are talking to AI or encountering AI content.[^1]

The EU wants to reduce deception and manipulation. They call these requirements transparency obligations. The EU AI Act is also known in German as the AI Regulation. It has been in force since August 2024 and is being implemented step by step.[^2]

Since when does the requirement apply?

Article 50 has been in effect since August 2, 2026.[^3] Since then, providers and deployers must fulfill the transparency obligations. You do not need to retroactively label content from before that date. However, the EU Commission recommends it where possible.

A limited transition period applies for providers of existing AI systems. It concerns only machine-readable marking and ends on December 2, 2026. Deployers who publish content do not benefit from this.

What risk classes does the EU AI Act recognize?

The EU AI Act categorizes AI systems according to their risk. The higher the risk, the stricter the rules. The model has four levels. They are called unacceptable, high, limited, and minimal risk.[^4]

Where does labeling fit in the risk model?

The transparency obligations from Article 50 are usually assigned to limited risk. Here, the EU does not require complex approval. It requires transparency toward people who come into contact with AI. The obligations apply in addition to the rules for high-risk systems.[^5]

What is prohibited and what is considered high-risk?

Unacceptable are AI applications that represent a clear threat to the safety and rights of people. High-risk systems are subject to strict requirements. This applies, for example, to applications in critical infrastructure.

For most marketing tools such as text and image generators, transparency obligations are central.

What does Article 50 require from providers and deployers?

Article 50 distributes the obligations across two roles. Providers develop AI systems and bring them to market. Deployers use these systems professionally and under their own responsibility.

Who is considered a provider and who a deployer?

The manufacturers of text and image generators are providers. Your marketing team is typically a deployer as soon as it uses AI tools for work.[^3] Individual employees are not considered separate deployers. The company bears the obligation.

The company remains the deployer if freelancers or service providers operate the tools on its behalf and under its responsibility and control.[^3] If you commission an agency only with the result and do not decide whether and how they use AI, according to the guidelines, the agency is the deployer. Anyone who only integrates a finished system also remains a deployer. Therefore, clarify responsibility for labeling contractually.

What obligations do deployers have?

Deployers must disclose two things. They label deepfakes. They also label AI texts that inform about topics of public interest, provided no human has editorially reviewed the text.[^3]

The labeling must be clearly visible. Users must perceive it at the latest upon first encountering the content. The provider's marking does not replace your visible labeling.

A separate information obligation applies for emotion recognition and biometric categorization. In typical marketing operations, it plays hardly any role.

What applies to machine-readable marking?

Providers of generative AI systems must mark their outputs in a machine-readable way. This allows audio, image, video, and text content to be recognized as AI-generated.[^5] This is the provider's responsibility, not yours.

Exceptions include, among others, standard editing, such as simple corrections.[^3] When selecting tools, check whether the provider guarantees the marking.

Which marketing content do you need to label?

Not every AI content needs a label. Content type, context, and human review are decisive. The following overview helps you with initial classification.

Content Labeling required? Reason
Realistic-looking AI image or AI video of a person, place, or event Probably yes The content may be considered a deepfake
Photorealistic stock image with fictional person Probably yes Even realistic, fictional persons can be considered deepfakes if viewers believe the image is real
Recognizably stylized AI graphic that no one would consider real Usually no No false impression of authenticity is created
AI text with human review and clear responsibility No The exception for editorial review applies
AI text on topics of public interest without editorial review Yes Article 50 paragraph 4 requires disclosure
Advertising text or product description Usually no Advertising is exempt, except for statements about health, consumer safety, or sustainability

In case of doubt, check each individual case. The EU Commission's guidelines provide examples for this. A non-binding initial assessment is also offered by the labeling check on our AI labeling page.

Does the AI labeling requirement apply to images and videos?

The AI labeling requirement for images and videos applies to a deepfake. This is AI-generated or AI-modified content that resembles real persons, objects, places, or events. Additionally, it must falsely appear authentic.[^3]

Context matters. The guidelines allow you to consider your audience's expectations. For artistic, creative, or satirical works, a notice that does not disturb enjoyment of the work is sufficient.

When in doubt, label rather once too many than too few.

Do you need to label AI texts?

Not every AI text requires labeling. The obligation only applies to texts you publish to inform the public about topics of public interest.[^3] These include, for example, health, consumer safety, or finance. Whether a blog article falls under this depends on the topic.

An exception applies with human review with clear editorial responsibility. A person must review the content professionally and substantively. A spell check is not sufficient. Additionally, someone must bear final legal responsibility for publication.[^3]

The number of contents is growing rapidly. This is evident in hyperpersonalization, when a language model generates texts for individual users. If AI agents even publish content independently, each step needs clear approval.

What applies to chatbots on the website?

Chatbots must inform people from the beginning that they are talking to AI. This does not apply if it is obvious.[^3] This obligation initially lies with the provider of the system.

You usually integrate an external tool. Therefore, check whether the provider includes the notice. If it's missing, add it yourself. You design the name, avatar, and greeting yourself anyway. A clear notice in the chat header or greeting is the right approach. This way users immediately know who they're writing to.

What applies to AI content on social media?

On social media, the same rules apply as everywhere else. What matters is whether a deepfake or AI text of public interest exists. Creators who regularly earn money with their posts are also considered deployers.[^3]

Decisions are made quickly in the feed. With social commerce, users often buy within seconds. Therefore, place the notice directly in the image or video. Users should perceive it without additional steps. You should not rely on platform labels, as they are lost when sharing or downloading.

How do you label AI content in practice?

The legislator does not prescribe a fixed form. However, two EU tools give you clear guidance.

What does the EU's Code of Practice offer?

The Code of Practice on transparency of AI-generated content is a voluntary guide. Independent experts developed it. It has two sections, one for providers and one for deployers.[^6]

Participation is voluntary. The obligations from Article 50 remain law nonetheless. The Commission and AI Board consider the code a suitable means to demonstrate compliance.[^6] Those who do not sign it must prove the suitability of their measures themselves. This risks more inquiries from authorities.[^3]

How do you use the official EU icons?

The EU has provided a set of EU icons. Deployers can use them to uniformly label AI-generated content.[^7] Place the icon so that users see it at first glance.

Use the icon consistently across all channels. This way your target audiences recognize the labeling more quickly.

How do you document AI use in the team?

Define who approves AI content. Document in writing which tools you use and how you review content. A four-eyes principle as in editorial planning can easily be linked to the required human review.

A clear content strategy also defines where you use AI and where humans work alone. Also name the person who bears editorial responsibility.

Who monitors compliance and what are the consequences of violations?

National market surveillance authorities primarily enforce the rules. The EU's AI Office only takes on a small part.[^3] In Germany, the implementation law for the AI Regulation designates the Federal Network Agency as the central market surveillance authority, unless another specialized authority is responsible.[^2]

Fines can amount to up to 15 million euros or 3 percent of worldwide annual revenue. For small and medium-sized enterprises, authorities consider proportionality.[^3]

Authorities also conduct technically precise reviews in data protection. What this looks like for consents is shown in our article on consent management platforms.

Frequently asked questions about the AI labeling requirement

Do I need to label every AI text?

No. The requirement only applies to AI texts on topics of public interest without human editorial review. If a person has professionally reviewed the text and bears responsibility, the requirement does not apply.

Is the AI tool's marking sufficient as labeling?

No. The provider's machine-readable marking does not replace your visible labeling. People must be able to perceive the notice themselves.

Do I need to retroactively label older content?

No. You do not need to retroactively label content from before August 2, 2026. However, the EU Commission recommends it where possible.

Does the requirement also apply to freelancers and service providers?

A company remains the deployer if freelancers or service providers use the AI on its behalf and under its control. If you commission an agency only with the result and do not decide whether and how they use AI, the agency is the deployer. Therefore, clearly regulate labeling in the contract.

Does the requirement apply to private social media posts?

No. Purely private use does not fall under the EU AI Act. However, anyone who regularly derives economic benefit or acts professionally is considered a deployer.

Conclusion

The AI labeling requirement does not demand blanket labeling from marketers, but clear processes. You visibly label deepfakes and certain AI texts. Human review with clear responsibility exempts texts from the requirement. The Code of Practice and EU icons provide you with concrete tools for this.

Start with an overview of your AI tools and content types. Then define who approves content and how you label. We can take care of the technical implementation of labels, metadata, and approvals in website, shop, and CMS for you. This article provides guidance and does not replace legal advice. Clarify individual cases with a specialist lawyer.

Would you like to align your content and processes with the AI labeling requirement without losing visibility? We're happy to discuss this with you in a free initial consultation.

[^1]: European Commission (2026): Guidelines on the transparency obligations for providers and deployers of certain AI systems (Article 50 AI Act). Available online at: https://digital-strategy.ec.europa.eu/de/policies/guidelines-transparency-ai-generated-content [^2]: German Bundestag (2026): Committee adopts law implementing the AI Regulation, hib 469/2026. Available online at: https://www.bundestag.de/presse/hib/kurzmeldungen-1184102 [^3]: European Commission (2026): Questions and answers on the transparency obligations under Article 50 of the AI Act. Available online at: https://digital-strategy.ec.europa.eu/de/faqs/transparency-obligations-under-article-50-ai-act [^4]: European Commission (n.d.): Regulatory framework proposal on Artificial Intelligence. Available online at: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai [^5]: AI Act Service Desk of the European Commission (n.d.): Article 50, Transparency obligations for providers and deployers of certain AI systems. Available online at: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50 [^6]: European Commission (2026): Code of Practice on Transparency of AI-generated Content. Available online at: https://digital-strategy.ec.europa.eu/de/policies/code-practice-ai-generated-content [^7]: European Commission (n.d.): EU icons for labeling AI-generated content. Available online at: https://digital-strategy.ec.europa.eu/de/policies/eu-icons-labelling-ai-generated-content