Do You Have to Disclose AI-Generated Content? EU Rules

A photo of a founding team that never stood together. A voice that sounds like the managing director but came from an AI. A local news summary nobody read before it went live. Since 2 August 2026, the EU requires disclosure for precisely these kinds of content.

For most small businesses, the obligation is narrower than the sales pitches for “AI Act compliance packages” suggest. Know where it applies and you can act where it matters and relax elsewhere.

What Article 50 of the AI Act covers

Article 50 of the AI Act contains several transparency duties. Some fall on the providers of AI tools, the likes of OpenAI or Google. Others fall on deployers, meaning businesses that use those tools.

Criterion	Obligation falls on	Applies from
Chatbots must identify as AI (para. 1)	system provider	2 Aug 2026
Machine-readable marking of AI output (para. 2)	generative AI provider	2 Aug 2026; legacy systems by 2 Dec 2026
Disclosing deepfakes (para. 4)	deployer, i.e. the user	2 Aug 2026
AI text on matters of public interest (para. 4)	deployer, i.e. the user	2 Aug 2026

Source: Regulation (EU) 2024/1689, Article 50; transition period per Regulation (EU) 2026/1744

Machine-readable marking is the providers’ job. Generate an image with a major AI tool and that tool should embed the marking itself, in metadata or as a watermark. Systems already on the market before 2 August 2026 have until 2 December 2026 under the Digital Omnibus.

As a user, the duties in paragraph 4 are the ones that concern you.

Deepfakes: the main obligation for businesses

The AI Act defines a deepfake as AI-generated or manipulated image, audio or video content that resembles real people, objects, places or events and would falsely appear authentic. Publish one, and you have to disclose that it was artificially generated or manipulated.

In practice:

• A photorealistic AI image of “our team at work” on your careers page is a candidate for a label.
• A video in which a real person says things they never said clearly is one.
• So is a cloned voice in an advert.
• An obviously drawn illustration, a cartoon or an abstract pattern isn’t, because none of them would pass as real.

For evidently artistic or satirical work, the duty is lighter: a notice that doesn’t spoil the display or enjoyment of the work is enough.

AI text: narrower than people think

For text, the obligation is much slimmer. It covers AI-generated or manipulated text published to inform the public on matters of public interest. News-like content, essentially.

There’s also an important exemption. The duty falls away when the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for the publication.

For a small business, that means newsletters, product copy or a specialist blog post that someone has read and approved don’t need a label. It gets tricky with text published automatically without human review, such as a bot summarising local news and posting it straight to a website.

1. Image, audio or video?Does it look real and could it pass as authentic? Then disclose it as AI-generated.
2. Art or satire?A discreet notice that doesn't spoil the work is enough.
3. Text?Does it inform the public on a matter of public interest?
4. Human-reviewed?With editorial responsibility, the text duty falls away.
5. Check platform rulesYouTube, Instagram and others have their own labelling rules.

Six everyday cases

The rules get clearer when applied to typical situations. These are our assessments based on the legal text, not advice on any specific case.

The last case is the grey zone. Minor retouching doesn’t make a deepfake. Once the product itself is altered, made smoother or larger or placed somewhere it never was, label it or don’t use it.

What a label can look like

The AI Act doesn’t prescribe a format, but the information must be clear and distinguishable at the latest at the first exposure. On 10 June 2026 the European Commission published a voluntary code of practice on marking and labelling AI-generated content. One section addresses providers, another deployers, and the Commission offers EU icons for labelling alongside it.

Wording that works in practice:

Image: "AI-generated" as a legible note in a corner of the image
       or directly in the caption
Video: on-screen notice at the start, "This video contains AI-generated scenes"
Audio: a spoken notice before the content
Text (where required): "This text was produced with AI" at the top

A notice buried in your terms or legal page won’t do. It has to appear where people encounter the content.

Rules beyond the AI Act

The AI Act isn’t the only rule in play. EU consumer law bans misleading commercial practices, so an AI image showing a result your product can’t deliver may be unlawful even if it isn’t a deepfake. Depicting real people in AI content also touches their personality and image rights.

Outside the EU, there’s no general labelling law in the UK or the US, but fake reviews are banned in both. The US Federal Trade Commission’s rule on fake reviews, in force since October 2024, explicitly covers AI-generated ones. In the UK, the Digital Markets, Competition and Consumers Act has banned fake consumer reviews since April 2025.

Platforms add their own layer. YouTube and Instagram expect creators to flag realistic AI content and apply labels of their own. When uploading there, use the disclosure option.

Fines: how serious is it?

For breaches of Article 50, Article 99(4) of the AI Act sets fines of up to 15 million euros or 3% of worldwide annual turnover, whichever is higher. For SMEs, paragraph 6 applies the lower of the two. Enforcement sits with national market surveillance authorities.

Realistically, regulators will start with cases that carry real potential to deceive, such as political deepfakes or fabricated statements by real people. For a small business that occasionally uses an AI image, a clean habit matters more than fear.

One rule for your team

The simplest approach is a house rule that belongs in your AI usage policy: photorealistic AI images, AI video and cloned voices are always labelled. Text is always reviewed and approved by a person. Those two sentences cover Article 50 for most companies and sit well alongside the AI literacy duty explained in our piece on AI Act AI literacy.

If you produce content with AI regularly, set up a content assistant that insists on sign-off; our walkthrough on building your own AI assistant shows how. And why human review makes sense beyond the legal angle is covered in our guide to AI hallucinations.

Anyone building a chatbot or content system for customers should plan disclosure and approval workflows from day one. Development partners such as OnlineBuilders build AI assistants for companies with that in mind.

Your next step: go through your website and social channels and flag every photorealistic AI image. More on law and risk is collected under privacy and limits.

Questions people ask

Do I need to label an AI-generated product image in my shop?

Under Article 50, only if it counts as a deepfake: realistic and liable to be taken as authentic. Separately, consumer protection law bans misleading advertising, so an AI image showing a product as better than it is can be a problem regardless.

Do I have to say a blog post was written with AI?

Usually not, if a person reviewed it and someone holds editorial responsibility. The text rule targets content published to inform the public on matters of public interest, and even there the editorial-control exemption applies.

How large are the fines?

Article 99 of the AI Act allows fines of up to 15 million euros or 3% of worldwide annual turnover for breaches of Article 50, whichever is higher. For small and medium-sized enterprises, the lower of the two applies.

Does this apply outside the EU?

The AI Act can reach companies outside the EU when their AI output is used in the Union. The UK and US have no equivalent general labelling law, but both ban fake reviews, including AI-written ones.