Labeling AI-generated text under the EU AI Act

The EU AI Act is Regulation (EU) 2024/1689, the European Union's horizontal law on artificial intelligence. Its Article 50 covers transparency, and it has applied since 2 August 2026. It binds anyone who publishes in the EU, including businesses based outside it that address an EU audience. Since August, the same question keeps arriving: do we now have to label every text that an AI helped write? The short answer is no.

This post is an overview of what the European Commission publishes itself. It is not legal advice, and for your own case it does not replace a lawyer's assessment.

What Article 50 regulates

Article 50 of the AI Act is about transparency, not quality. It exists so that people can tell when they are dealing with artificially generated content. It applies from 2 August 2026.¹

The law separates providers from deployers. A deployer is the entity using an AI system under its own authority, and the Commission names an advertising company as an example.¹ Individual employees acting on that entity's instructions and under its control are not separate deployers, and the entity stays the deployer even when contractors or freelancers are involved in operating the system on its behalf.¹

The penalties explain part of the nervousness: fines can reach 15 million euros or 3 percent of total worldwide annual turnover, with proportionality for small and medium-sized enterprises.¹

When AI-written text has to be labeled

For text, the Commission names three conditions that must apply together. The text is published, it informs the public, and it concerns a matter of public interest.¹ The law itself speaks of deployers of an AI system "that generates or manipulates text which is published with the purpose of informing the public on matters of public interest".²

For a small business, the middle rows matter most. A service page drafted with AI help does not fall under this duty. A blog post about the legal situation, an industry development or a new technology can.

The exception: review plus editorial responsibility

Article 50(4) contains the exception that decides most cases. The duty does not apply "where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content".²

Two conditions, not one. There must be a real review, and someone must carry the responsibility. The Commission describes editorial responsibility as holding the ultimate legal responsibility over the publication of the content.¹

It is equally clear about what does not count: "Superficial, solely formal, or procedural checks (e.g. spell-checking or grammatical correction) are not considered to be human review or editorial control."¹ Running a spelling checker over an unread draft does not meet the exception.

What this means for your website and social posts

A workable order of steps:

1. Ask whether the text informs the public on a matter of public interest at all. For service pages, product texts and ads, the answer is usually no.
2. If yes, make sure a person actually reviews the content, beyond spelling and grammar.
3. Record who holds editorial responsibility. A name on the page or in the site's legal notice is enough for this purpose.
4. If neither the review nor the responsibility is in place, label the text as AI-generated.

Example

Content generated before 2 August 2026 does not need to be labeled retroactively, although the Commission encourages it where possible.¹ The machine-readable marking duty for providers, meaning the technical signature inside the content, has a transition period until 2 December 2026, and only for systems placed on the market before 2 August 2026.¹

Images, audio and video follow a stricter rule

For deepfakes, meaning artificially generated or manipulated image, audio or video content that appears authentic, there is no comparable editorial exception. Deployers must disclose such content to a person upon first exposure at the latest, in a perceivable way, for example with visible or audible labels.¹ The law requires the information "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure".²

One addition from the Commission matters in practice: deployers cannot simply rely on the machine-readable marking that the provider embedded in the content.¹ An invisible signature inside the file does not replace a label the viewer can see.

For a restaurant, that is concrete. A generated but plausible photo of your own terrace needs a label. An obviously drawn illustration is a different case.

Platform rules can be stricter than the law

The law is not the only constraint. Since 2 August 2026, Google does not allow AI-edited or AI-generated images in Local Services Ads in the European Union.³ There, no label helps, because the material itself is not allowed.

So check the platform policy of the channel you publish on first, then the legal duty, and follow whichever is stricter.

What this means for your business

For your own business, we suggest this order: name a person, read texts before publishing, and label when in doubt.

Notes

1. European Commission, "Transparency obligations under Article 50 of the AI Act", last updated 24 July 2026, accessed 16 October 2026. ^{2 3 4 5 6 7 8 9 10 11}

2. Article 50 of Regulation (EU) 2024/1689, read on artificialintelligenceact.eu, accessed 16 October 2026. The official text on EUR-Lex was not reachable that day, so the wording was verified against this unofficial copy. For legal work, the Official Journal text governs. ^{2 3}

3. Google Local Services Help, "Local Services Ads requirements", accessed 16 October 2026. The rule also covers the US state of New York and, according to the page, does not apply to advertising through partner affiliates.