AI Act: What EU Rules Change for Brands' Content

Since August 2, 2026, Article 50 of the EU Artificial Intelligence Act (AI Act) has imposed two obligations that directly concern brands: telling users clearly that they are talking to an AI, and labeling certain content (text, image, audio, video) generated or manipulated by AI. This isn't a marketing choice, it's a legal obligation backed by fines. Here is what the text says, since when, for whom, and what is still being decided in France.

This article relies only on the text of Regulation (EU) 2024/1689, on official European Commission pages, and on French legislative texts still under discussion. It is not legal advice: for a precise compliance review, a specialized lawyer remains necessary.

What is the AI Act, and when does it apply?

The AI Act is Regulation (EU) 2024/1689, which entered into force on August 1, 2024, and governs artificial intelligence systems across the European Union. It doesn't apply all at once: its obligations take effect in stages, and the August 2, 2026 stage is by far the widest.

The schedule, according to the European Commission's official implementation timeline:

• February 2, 2025: definitions, AI literacy duties, and the prohibited practices under Article 5 start applying.
• August 2, 2025: rules for general-purpose AI models take effect, and member states must have their national governance in place.
• August 2, 2026: most of the regulation applies, including the high-risk rules under Annex III and the transparency rules under Article 50. Enforcement begins.
• August 2, 2027: rules for high-risk AI embedded in products already regulated elsewhere (medical devices, toys, machinery…) apply.

August 2, 2026 is the date that matters for nearly every brand: it is the one that brings Article 50 into force, the only article aimed at almost every company using an AI tool, not just those building high-risk systems.

The AI Act is a compliance text, not a visibility lever: it says nothing about how an AI engine decides to cite your brand in its answers. That separate question, GEO, is covered in our comparison between SEO and GEO and in our complete guide to GEO. Compliance and visibility are independent: a site can be fully compliant with the AI Act and still be invisible in ChatGPT's answers, or the reverse.

What does Article 50 change for a chatbot on your site?

It must clearly tell the user they are talking to an AI, unless this is already obvious to a reasonably well-informed, observant and circumspect person.

The official text of Article 50(1), as reproduced on the European Commission's AI Act Service Desk, requires providers of AI systems intended to interact directly with natural persons to design them so that those persons are informed that they are interacting with an AI system, "unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use." The Commission's own FAQ states that this exception must be interpreted restrictively, since it deprives the user of information: it only applies when almost no doubt remains.

Two examples show the difference. A code-review tool available only to professional developers may fall under the exception: the context makes AI use obvious. A customer-support chatbot shown in a generic chat bubble on a consumer-facing site generally does not: a visitor can reasonably believe they're talking to a person. This is exactly the case we already flagged for bank and insurance chatbots, in our article on AI visibility in banking and insurance.

In practice, this means a clear notice on the very first interaction ("You are chatting with an AI assistant"), not just a line buried in a legal notice or terms of service.

Does AI-generated or AI-edited content need to be labeled?

Yes, in two specific cases: synthetic content (text, image, audio, video) produced by an AI system must be technically marked as such, and certain content published to inform the public on matters of public interest must be disclosed as AI-generated if it was not subject to human editorial review.

Article 50 sets out two distinct duties, aimed at different parties. Paragraph 2 addresses providers of generative AI systems: they must ensure that synthetic outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. Paragraph 4 addresses deployers, meaning companies that use a third-party AI tool: they must disclose that a text published to inform the public on matters of public interest was generated or manipulated by AI, unless that text has undergone a human review process and a natural or legal person holds editorial responsibility for its publication.

This second nuance matters most for a marketing or communications team: a blog post written with AI assistance but reviewed, checked and approved by a named writer falls outside this specific labeling duty. A text generated and published without any human review, on a matter of public interest, stays within it. The line between the two rests on the reality of editorial control, not a statement of principle.

For deepfakes (synthetic image, audio or video that convincingly depicts real people, objects, places or events in a misleading way), the disclosure duty also falls on the deployer, with a lighter exception for works that are evidently artistic, creative, satirical or fictional: an appropriate notice is then enough, without having to hide the nature of the work.

One important timing detail: the technical marking duty under paragraph 2 only applies from December 2, 2026 for generative AI systems already on the market before August 2, 2026. This four-month delay, agreed as part of the "Digital Omnibus" package on AI, only concerns this specific technical marking duty: chatbot disclosure (paragraph 1) and the labeling of deepfakes and public-interest text (paragraph 4) have applied since August 2, 2026.

Who enforces these obligations in France?

The split is not yet definitively set in French law: a bill designating competent authorities is still being examined by Parliament, and the European Commission has published non-binding guidelines to help apply the text.

At EU level, the Commission adopted guidelines on the implementation of Article 50 on July 20, 2026, following a public consultation on a draft published in May 2026. These guidelines are not legally binding: they help interpret the text rather than replace it, and should be read alongside the voluntary code of practice on the transparency of AI-generated content.

In France, the Direction générale des entreprises (DGE) and the Direction générale de la concurrence, de la consommation et de la répression des fraudes (DGCCRF) have published a draft plan splitting authority among existing regulators, rather than creating a single dedicated AI authority. An amendment to the bill adapting French law to EU law, tabled in the Senate and discussed in February 2026, would designate the CNIL, the DGCCRF, the DGE, Arcom and Arcep, coordinating their powers depending on the type of AI system involved. As of this article's publication, that text had not been definitively adopted: the exact split of oversight (whether the DGCCRF, Arcom or the CNIL reviews a given type of content or chatbot) may still change. A brand that wants to check the text's status can follow the Senate's and the DGE's publications.

What does a brand risk by ignoring these obligations?

An investigation by the competent authority, then, if a breach of Article 50 is confirmed, a fine of up to €15 million or 3% of worldwide annual turnover, whichever is higher.

The regulation sets several tiers of penalties in Article 99. The highest tier, up to €35 million or 7% of worldwide turnover, covers the prohibited practices under Article 5 (manipulation, exploitation of vulnerabilities, social scoring…), not Article 50. Breaches of Article 50 fall under an intermediate tier, set at €15 million or 3% of worldwide turnover, whichever is higher. For SMEs, including startups, Article 99 reverses that rule: the lower of the two amounts applies instead.

These penalties are separate from GDPR fines: a chatbot that collects personal data without a legal basis can face both a GDPR breach and an AI Act Article 50 breach, reviewed by authorities that are not necessarily the same. This is a legal matter, not a marketing optimization: if you're unsure of your own exposure, a legal review of your AI tools (chatbot, content generator, scoring tool) remains the only reliable way to know where you stand.

What should you actually do?

Five checks give you a first status report, without waiting for the final word on the French authorities.

1. List every AI system that interacts with your visitors or customers: support chatbot, voice assistant, sales-qualification agent, conversational recommendation tool.
2. Check that each one displays a clear notice on the very first interaction, not just in a legal notice. "You're chatting with an AI assistant" is enough in most cases.
3. Identify content published without human review: product sheets, summaries, automated answers published on matters of public interest. These remain exposed to the labeling duty even with the technical-marking delay to December 2026.
4. Put in place, or document, a real human editorial review of AI-assisted content: a named writer who checks and approves before publication takes most text out of the labeling duty's scope.
5. Track how the designation of French authorities evolves before picking a contact for questions: it isn't settled as of this article.

These obligations govern how you inform your visitors, not how AI engines talk about you to your prospects. That second question, visibility, is measured separately: our guide to measuring AI visibility compares the available methods. For a full diagnosis across four AI engines (ChatGPT, Gemini, Claude and Perplexity) and what they say about your brand against competitors, Beeleven's AI visibility audit is the right tool — a separate question from regulatory compliance, but one that often comes up at the same time in marketing departments.

And if an AI answer says something inaccurate about your brand, the AI Act doesn't fix that: it's an AI reputation issue, addressed through the sources AI engines consult, not through a compliance notice. In the same way, if you're considering advertising inside ChatGPT, the "sponsored" labeling we describe in our article on ChatGPT Ads in France follows a commercial logic, different from the AI Act's transparency duty, even though both aim to let the user know what they're looking at.

FAQ

Does the AI Act apply to every company?

Article 50's obligations apply to any company using an AI system that interacts directly with people, or that generates synthetic content, regardless of size. Only the scale of penalties changes for SMEs, with a lower cap under Article 99.

Does my chatbot need to say it's an AI even when that's obvious?

No, Article 50 includes an exception when this leaves no doubt for a reasonably well-informed, observant and circumspect person, given the context. But the European Commission's FAQ states that this exception must be read restrictively: when in doubt, show the notice.

Do I need to label an article written with AI help?

Only if it's text published to inform the public on a matter of public interest and it hasn't undergone genuine human editorial review. An article checked and approved by a named writer falls outside this specific Article 50(4) duty.

Which authority enforces these obligations in France?

This isn't definitively settled by law yet: a bill under discussion would split enforcement between the CNIL, the DGCCRF, the DGE, Arcom and Arcep depending on the type of AI system. The text had not been definitively adopted as of this article's publication.

Sources

• European Commission, AI Act Service Desk, Article 50 text on transparency obligations.
• European Commission, AI Act Service Desk, Article 99 text on penalties.
• European Commission, AI Act Service Desk, implementation timeline.
• European Commission, FAQ on Article 50 transparency obligations.
• European Commission, draft guidelines on the implementation of Article 50, published May 2026, adopted July 20, 2026.
• Direction générale des entreprises, press release on the draft designation of national AI Act authorities.
• French Senate, amendment No. 284 to the bill adapting French law to EU law, February 2026.
• Blog du Modérateur, "IA Act: ce qui change le 2 août 2026" (French).

→The complete guide to GEO: understanding AI visibility beyond compliance

→GEO audit: how to audit your AI visibility