Do you have to label your articles as AI-generated? What article 50 of the AI Act says

What article 50 actually requires

Article 50 of the AI Act, the European regulation on artificial intelligence, is mostly about transparency. It does not ban AI-produced content: it requires that users know when they are interacting with a machine, or when content has been artificially generated or altered.

The European text separates several situations. Some call for explicit information to the user or the public, others carry a closed list of exceptions and adjustments: use authorised by law for law enforcement purposes, the adjustment provided for manifestly artistic works, and human review or editorial control under assumed editorial responsibility. In practice the question is therefore not "was AI used?" but rather: in which context is transparency mandatory?

Article 50 applies to very concrete uses: chatbots, voice assistants, synthetic content, deepfakes, systems that generate or manipulate image, audio or video, and certain interactions with automated systems. The AI Act is a European regulation: the obligations fall on those who make such systems or content available in the European Union.

The content that has to be flagged as AI-generated or manipulated

The clearest case concerns deepfake content. Where image, video or audio content is generated or heavily manipulated by AI and gives the impression of a real event, transparency becomes mandatory. The public has to be told the content was artificially generated or manipulated, subject to the exceptions the regulation provides.

Article 50 also covers AI-produced text published to inform the public on matters of public interest, along with audio, image and video content presented as authentic. The threshold is the risk of confusion: the more the content resembles credible human work, the more labeling is called for.

• Audio deepfakes: cloned voices, synthetic voice messages, fake calls.
• Video deepfakes: a face, gestures or a scene altered by AI.
• AI-generated images: portraits, news scenes, realistic visuals.
• Synthetic text: only where it is published to inform the public on matters of public interest. That is the criterion, not how sensitive the context is: commercial copy falls outside the scope even if it could be taken for human writing, as long as it carries no health, consumer-safety or sustainability claim.

The obligation does not always mean displaying a uniform "AI-generated" line on every piece. The rule depends on the distribution context, the risk of confusion and the nature of the content. An advertisement, a news item, a testimonial and a conversational avatar are not assessed the same way.

Interactions with AI systems also have to be transparent

Article 50 also covers human-machine interaction. Where someone communicates with an AI system in conditions where they might believe they are talking to a person, they have to be told they are interacting with a machine.

That requirement covers in particular:

1. customer service chatbots;
2. voice assistants;
3. conversational systems built into sites, apps or messaging;
4. automated agents used in sales or support journeys.

The transparency notice can sit at the start of the exchange, in the interface, or take the form of a clear and intelligible signal. The user has to know who they are talking to. The AI Act gives that practice a European legal basis.

One useful clarification, because the confusion is common: article 50 creates no right to obtain human intervention. It deals with transparency alone, that is, with knowing that the service is automated. The right to ask for human intervention against an automated decision comes from article 22 of the GDPR, and the right to an explanation of a decision taken with a high-risk AI system comes from article 86 of the AI Act. For brands and publishers, article 50 therefore means reviewing conversation journeys, scripts, landing pages and the legal notices attached to conversational tools, not opening a human channel.

The exceptions the AI Act provides

Article 50 does not impose systematic labeling in every case. The regulation provides adjustments and exemptions, and they are not equivalent: some lighten the form of the disclosure, others remove it.

It is the second row that matters for a publisher, and it is broader than it is usually read. The regulation sets out two alternative routes: human review or editorial control. Reviewing every article is therefore not the only path; editorial control exercised over production opens the same right, on the cumulative condition that a natural or legal person holds editorial responsibility for the publication. We set out this framework in our AI Act timeline for a publisher, and the byline that embodies it in one article, one signatory. What each of the two routes requires, and what voids the exemption, is the subject of a dedicated article on the exemption.

These exceptions show that the European text seeks a balance between transparency and creative freedom. A film, an artistic illustration or a parody are not treated like a viral video claiming to show a real event. The risk of deception is the compass.

For content professionals, that means analyzing a piece's purpose, format, destination and degree of realism. Simple AI assistance in drafting an article does not automatically trigger an obligation to say "AI-generated" if the finished result does not try to pass for anything other than ordinary editorial content.

How to apply article 50 in practice without over-labeling

The best approach is to think in terms of useful disclosure, not automatic labels everywhere. A useful disclosure appears where the risk is, and is understood without a glossary. In practice, companies can put several levels of information in place:

• on the content itself: a line visible directly on the image, the video, the audio or the page;
• in the interface: a banner or a label inside the conversational tool;
• in the editorial policy: a page explaining when and how AI is used;
• in the metadata: useful for internal organization, though it does not always replace information visible to the user.

For marketing, press, e-commerce or customer relations teams, these questions frame the need:

1. Can the content be mistaken for the real thing?
2. Does the audience expect human authenticity?
3. Does the context make the disclosure necessary to avoid deception?

If the answer to any of these is yes, the disclosure is required. Displayed where it is useful, it also becomes an argument for the brand's credibility.

Key points, and frequently asked questions

Article 50 of the AI Act mainly targets synthetic content liable to deceive and interactions with AI systems. The content most concerned is deepfakes, images, audio recordings and video generated or manipulated by AI, and exchanges with chatbots or assistants where the user might believe they are talking to a person.

The obligation is therefore targeted. It imposes a logic of transparency fitted to the context, and it falls away for texts of public interest that have undergone human review or editorial control under assumed editorial responsibility. For publishers, brands and content producers, what remains is to document how AI is used and to place clear disclosures wherever the risk of confusion is real.

• Does every piece created with an AI have to be marked? No. The obligation depends on the type of content, the context and the risk of deception.
• Does a reviewed and signed article have to carry the disclosure? No, provided the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication. You still have to be able to prove it: who reviews, at which stage, who signs, and where the record is kept.
• Do AI-generated images always have to be flagged? They do where they can be taken for real images or where they carry misleading information.
• Does a chatbot have to announce that it is an AI? Yes, where the user might believe they are dealing with a person.
• Is artistic content covered? Yes, and this is not an exemption: where a deepfake is part of a manifestly artistic, creative, satirical or fictional work, the obligation is limited to disclosing the existence of the generated content in a way that does not hamper the display or enjoyment of the work.
• Does the AI Act only concern content visible in Europe? It covers systems and content made available in the European Union, which calls for particular care from international players.