Labelling AI-Generated Content from 2 August 2026 – what you actually have to label in the EU

EU rules on artificial intelligence do not introduce a blanket obligation to label every piece of material created with the help of AI. In practice, what matters are specific transparency obligations under Article 50 of the AI Act, which apply to particular types of systems and particular categories of content — not the entire internet or all marketing.

This distinction is important, because public debate often conflates three different levels: the mere fact that an AI tool was used, the legal obligation to label certain outputs, and compliance best practices that a company may implement more broadly than the law requires. If you want to prepare a website, store, portal, newsroom or content team for the new obligations, you need to start with a correct understanding of the scope of the rules.

What is the legal basis

The regulatory basis is the EU Artificial Intelligence Act, i.e. the AI Act, presented by the European Commission as the first comprehensive EU legal framework for AI. In the area of content labelling, the key provision is Article 50, which establishes transparency obligations for certain AI systems and certain uses of generative AI.

The European Commission states explicitly that the obligations under Article 50 apply to providers and deployers of generative AI systems and cover the marking and detectability of AI-generated content, the labelling of deepfakes, and certain text publications. In parallel, on 10 June 2026 the Commission published a voluntary Code of Practice on the marking and labelling of AI-generated content, intended to help demonstrate compliance with these obligations.

Official sources:

• AI Act – European Commission
• Code of Practice on marking and labelling of AI-generated content – European Commission
• AI Act Service Desk FAQ

When the obligations apply from

The European Commission states that the transparency obligations concerning the marking and labelling of AI-generated content, arising from Article 50, apply from 2 August 2026. This date appears explicitly on the official Commission page dedicated to the Code of Practice.

The Commission page also indicates that these obligations complement other parts of the AI Act, such as the rules on high-risk systems or general-purpose AI models. In other words, the labelling of AI content is not a separate law, but part of the broader AI Act regulatory system.

Who the obligations apply to

The Commission splits the obligations into two basic groups: providers and deployers of generative AI systems. This distinction is crucial, because a manufacturer or provider of a model or tool has different obligations than a company, newsroom or store that uses such a system to publish content.

1) Providers of AI systems

According to the Commission’s materials, the section for providers concerns the rules for marking and detecting content generated and manipulated by AI. The Commission indicates that the outputs of AI systems — including audio, image, video and text — should be marked in a machine-readable format and detectable as artificially generated or manipulated.

The Commission page adds that the technical solutions used should be effective, interoperable, robust and reliable to the extent that this is technically feasible. This means that the burden of the technical layer rests primarily on the providers of generative tools, and not solely on the end user publishing the content.

2) Deployers

The section for deployers concerns the disclosure of artificially generated or manipulated content that constitutes a deepfake, and of text publications generated or manipulated by AI that inform the public on matters of public interest. The Commission also points to an exception: the obligation for such text publications does not apply where the publication has undergone a process of human review and is subject to editorial responsibility.

It is precisely this passage that carries the greatest significance for portals, blogs, newsrooms, publishers and companies publishing informational content. In practice, then, it is not about every piece of marketing content or every product entry, but about a narrower area defined by the provision and clarified by the Commission.

What content has to be labelled

Content generated or manipulated by AI

The Commission indicates that the transparency obligations cover content generated by AI as well as content manipulated by AI. Audio, image, video and text are explicitly listed.

This is important, because the rules are not limited solely to classic “generator graphics” or “deepfake videos”. The scope is broader and also covers the manipulation of existing content, where an AI system changes its character in a way that is material from the recipient’s point of view.

Deepfakes

The Commission defines a deepfake in this context as an image, sound or video resembling existing persons, objects, places, entities or events, which would falsely appear to a recipient to be authentic or truthful. If a deployer publishes such content, it has an obligation to disclose that it is artificially generated or manipulated.

For business, this means particular risk in video materials, synthetic voice statements, avatars, image-based advertising, and content styled to look like an authentic recording of a real person or event. The greater the resemblance to reality and the greater the potential to mislead the recipient, the more important correct labelling becomes.

Text informing on matters of public interest

The Commission indicates that the disclosure obligation also covers text publications generated by AI or manipulated by AI which serve to inform the public on matters of public interest. At the same time, the Commission reserves an exception for materials that have undergone a process of human review and are subject to editorial responsibility.

This is one of the most important points of the entire regulation, because it shows that the EU obligation is not written as a universal mandate to label every text created “with the help of AI”. If a piece is a publication on matters of public interest but has undergone genuine human control and editorial responsibility exists, the scope of the obligation looks different than for a purely automated publication.

Interaction with an AI system

The materials describing Article 50 also give rise to an obligation to inform users that they are interacting with an AI system, for example a chatbot, unless this is obvious from the circumstances and context of use. This matters for websites, e-commerce, customer service, lead generation and all conversational AI deployments.

So if an assistant, voicebot or conversational widget operates on a site, the user should receive clear information that they are communicating with an AI system. In practice, it is safest to do this right at first contact, within the conversation interface itself.

What the rules do not say explicitly

The rules and the official Commission materials do not confirm the claim that every text written using AI must always be labelled for the end recipient. The official Commission page narrows the obligations of text deployers to publications informing the public on matters of public interest, excluding materials that have undergone human review and are under editorial responsibility.

This means that in ordinary content marketing, category descriptions, product descriptions, landing pages or sales articles, you must not automatically assume an obligation of public labelling solely because AI helped with the editing. At the same time, from a compliance and reputational-risk perspective, many companies may nonetheless choose to adopt broader internal standards than the required minimum.

The role of the Code of Practice of 10 June 2026

The Code of Practice published by the Commission on 10 June 2026 is voluntary in nature. The Commission emphasises clearly, however, that the transparency obligations under Article 50 of the AI Act themselves are legally binding.

The Code is meant to help providers and deployers demonstrate, in practical terms, compliance with the obligations concerning the labelling and marking of AI-generated content. The Commission also indicates that signatories of the Code will be able to rely on the measures described in it as a practical compliance framework recognised across the EU.

This means a simple thing: the Code does not replace the law, but it can be a very useful implementation standard. For companies and publishers this is important, because it allows compliance processes to be built according to an officially promoted template, rather than creating one’s own interpretations from scratch.

What labelling should look like in practice

The official Commission materials show two levels of labelling. The first is the technical level, in which content generated or manipulated by AI is to be marked in a machine-readable format and detectable as artificial.

The second level is the communication layer for the recipient, especially important for deepfakes and for certain text publications and interactions with AI. The Commission also states that it has prepared a set of EU icons for labelling AI-generated content, which deployers can use in practice.

For companies, the most reasonable approach is therefore a two-layer model:

• technical marking where this applies to the provider of the system or the publishing infrastructure;
• human-readable labelling where the rules require disclosure to the recipient;
• documenting the human review process where the company wants to demonstrate that the publication was not an automated publication without editorial responsibility.

What this means for different types of sites and businesses

Portal, media, news service

The rules have the greatest significance for entities publishing materials on matters of public interest, especially where part of the editorial process is automated. If texts are generated or substantially processed by AI and serve to inform the public, you need to analyse whether the exception related to human review and editorial responsibility applies.

E-commerce

In e-commerce, the central importance lies rather in the layer of chatbots, voicebots, synthetic advertising materials and potential deepfakes, rather than in the mere fact of using AI to draft product descriptions. If a store uses conversational AI for customer service, the disclosure about interaction with AI should be implemented clearly and without hiding this information in the terms and conditions.

Marketing and SEO agency

For agencies, the key is to separate: what is merely editorial support, what is public-interest material, what is synthetic media, and what is a user’s interaction with AI. The biggest compliance mistake is applying one simple assumption to everything — for example “we label every AI text” or, conversely, “we don’t label anything, because it’s just editorial help”.

Implementation checklist for a company

1) Inventory your AI usage

First, list all the places where the company uses AI to create, rework or publish content. Separately mark: text, image, audio, video, chatbots, voicebots, and tools for editing or manipulating existing materials.

2) Separate the roles: provider vs deployer

Check whether your company is only a user of an external tool, or whether it also itself provides an AI function to clients or users. This determines whether you are responsible only for labelling towards the user, or also for the technical layer of marking and detectability.

3) Flag high communication-risk areas

Put into a separate category everything that could be perceived as authentic real material: avatars, synthetic voice, “recordings” of people’s statements, advertising styled as event coverage, news materials, and chatbots without explicit disclosure. This is precisely where the risk of misleading the recipient is greatest.

4) Build a publication policy

In your internal policy, establish when a piece must receive a public label, when editorial documentation is sufficient, who approves publication, and how the human review is documented. If you publish content on matters of public interest, it is worth describing separately the standard of editorial responsibility and the approval path.

5) Add disclosure to chatbots and AI interfaces

On sites and in apps, place a clear message that the user is talking to an AI, where this does not follow unambiguously from context. It is best to implement it at first contact, and not only in the terms and conditions or the footer.

6) Assess using the EU icons

The Commission has made available a set of EU icons for labelling AI-generated content. In practice, it is worth checking whether you can use them as an additional visual standard in materials where labelling is required or justified.

The most common interpretation mistakes

The most common mistake is equating “the use of AI in the creation process” with “an obligation to label every publication”. The official Commission materials show a much more precise scope of obligations.

The second mistake is ignoring the distinction between the provider of a system and the deployer. The third mistake is failing to separate text content from deepfakes and interactive AI systems, even though each of these categories has a different profile of obligations and risk.

The fourth mistake is basing compliance solely on a general disclaimer in the terms and conditions. In the areas covered by Article 50, what counts is real transparency towards the user and an appropriate labelling model — not merely hiding general information in legal documentation.

Practical conclusions

From a business perspective, from 2 August 2026 the point is not to “label everything”, but to correctly recognise what is subject to the obligation, in what role your company operates, and in what form transparency must be demonstrated. For most companies, the best solution will be a combination of an editorial policy, disclosure for AI interactions, a procedure for deepfakes, and documentation of human review for sensitive publications.

If you run a portal, store, agency or SaaS and use generative AI, it is worth treating Article 50 of the AI Act not as a minor formality, but as an element of governance, editorial responsibility and communication-risk management.

Sources to link under the article

The sources below are official or governmental and you can safely attach them under the publication as a source base:

• European Commission: AI Act
• European Commission: Code of Practice on marking and labelling of AI-generated content
• European Commission: AI Act Service Desk – FAQ