The EU AI Act (Regulation 2024/1689) entered into force on 1 August 2024 and applies in phases. For marketers the key date is 2 August 2026: from then on the transparency obligations of Article 50 apply to anyone who uses AI to create or show content. The regulation does not ban AI content and does not require a label on every text. It asks three things: be honest when people talk to an AI, make synthetic media recognisable, and do not publish AI text on matters of public interest without a disclosure or human editing. This article explains what that means concretely and how to make it demonstrable.
First the roles: provider and deployer
The AI Act places duties on two parties. The provider develops an AI system or model and places it on the market: the builder of the language model, or the vendor of a tool that runs on it. The deployer uses an AI system under its own authority: the company that produces text, images or chatbots with that tool.
A marketing team is almost always a deployer. That is good news and an assignment at the same time: the heaviest technical duties sit with the provider, but what you publish is your responsibility, and Article 50(4) is aimed precisely at you.
What Article 50 asks
Article 50 contains four obligations relevant to content.
Paragraph 1: people know they are talking to an AI. A chatbot on your site must make clear that it is an AI system, unless that is obvious from the context. This sits with the provider of the system, but if you run a chatbot under your own brand, it is your page where that has to be visible.
Paragraph 2: synthetic output is machine-detectable. Providers of systems that generate text, image, audio or video must mark their output in a machine-readable format and make it detectable as artificially generated. Think of watermarks and provenance metadata such as C2PA credentials for images. For you as a deployer this means: choose tools that support this and do not strip that metadata when publishing.
Paragraph 4, first part: deep fakes are labelled. Anyone who uses AI to generate or manipulate image, audio or video that constitutes a deep fake must disclose that the content has been artificially generated. For marketing this means a generated "customer portrait" or a manipulated video cannot go out without a notice.
Paragraph 4, second part: AI text on matters of public interest. This is the provision most questions are about. Anyone who publishes AI-generated or -manipulated text with the purpose of informing the public on matters of public interest must disclose that the text was artificially generated. The exception: this is not required where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for the publication.
That exception is the key for content teams. A product page or a comparison article that an editor has checked and that appears under your company's responsibility falls under it. A fully automated stream of texts without a human eye does not, and on "matters of public interest" (health, finance, law, safety, politics) you need to be careful in any case.
What this asks in practice
The regulation prescribes no workflow, but the combination of Article 50 and the ordinary burden of proof leads to four practical requirements.
- A review process that is demonstrable. Not "we proofread everything", but recorded per publication: who reviewed, when, and what changed. Without that evidence the exception in paragraph 4 is a claim.
- Provenance per page. Which model produced the text, with which instruction (a hash is enough), which version was published. This is the same record you need to trace where a mistake came from.
- A disclosure policy. Decide per content type whether you show a notice. For informative content about health, money or law a short, honest sentence ("drafted with the help of AI and reviewed by an editor") is sensible even where the exception applies. Put that choice on a public page, so a reader and a regulator can see how you use AI.
- Images with credentials. For generated images, use tools that attach C2PA metadata and make sure your CMS and CDN do not strip it. For deep-fake-like material: never without a label, and preferably not at all.
The European Commission is working on a code of practice for marking and labelling AI content that will make these provisions more concrete. If you arrange the four points above now, you do not have to wait for it.
What it does not ask
A few misunderstandings we often encounter. The AI Act does not require a label on every AI text; the duty is targeted and has an editorial exception. It does not ban AI content, not even in advertising; the ordinary consumer-law rules on misleading practices apply there. And it does not turn a marketing tool into a high-risk system; the high-risk categories concern things like recruitment, credit scoring and biometrics. What remains is the honesty standard: do not mislead people about what a machine made and who signs off on it.
Timeline to remember
- 1 August 2024: regulation in force.
- 2 February 2025: prohibited practices and the AI-literacy duty apply.
- 2 August 2025: obligations for providers of general-purpose AI models.
- 2 August 2026: Article 50 (transparency) and most of the remaining obligations apply.
- 2 August 2027: remaining high-risk obligations for AI in regulated products.
Conclusion
For marketers the AI Act comes down to a workable standard: let people know when they are dealing with AI, keep synthetic media recognisable, and publish AI text only under demonstrable editorial responsibility. If you record per page which model, which reviewer and which changes belonged to it, you have both the Article 50 exception and a better content workflow. See how AI content creation in Traze records every version with model, prompt hash and reviewer and switches a disclosure block on or off per site.