Labelling AI-generated content: what the EU AI Act requires of brands

Do we have to label AI-generated images and text in our marketing?

Labelling AI-generated content is the practice of telling people, through a visible notice or a machine-readable mark, that an image, video, audio clip or text was created or altered by AI. In the EU, parts of it became a legal duty on 2 August 2026, when Article 50 of the AI Act started to apply (Bird & Bird, 2026).

The short answer is that you do not have to label everything, but you have to label more than many brands assume. Article 50(4) gives two labelling duties to deployers, meaning the companies that use AI tools. Image, audio and video that counts as a deepfake must be disclosed as AI-generated or manipulated. AI-generated text must be disclosed when it is published to inform the public on matters of public interest, unless a person has reviewed it and someone holds editorial responsibility for it.

The Act defines a deepfake as AI-generated or manipulated image, audio or video content that resembles existing people, objects, places or events and would falsely appear authentic (Regulation (EU) 2024/1689, Article 3). The definition says nothing about intent to deceive. What counts is whether people would take the content for real.

In marketing, these cases sit close to the line:

• a photorealistic synthetic spokesperson

• a cloned voice in a radio or audio ad

• an AI-generated, real-looking setting around a real product

• a documentary-style image in which an event has been altered

A clearly illustrated or stylised invented world does not. Text is never a deepfake. AI-generated text falls under the separate duty for public-interest text.

Conclusion: The trigger is the risk of deception, not the use of AI.

Why do the EU AI Act transparency rules matter for brands?

The rules matter because they already apply, they carry real fines and they cover the tools marketing teams use every day. The European Commission says the transparency obligations help people recognise when content has been generated or altered by AI, "reducing the risk of deception and manipulation" (European Commission, 2026).

Fines can reach the greater of EUR 15 million and 3 per cent of worldwide turnover (Paul, Weiss, 2026). The rules were not delayed either. The EU's Digital Omnibus postponed the high-risk requirements, but the Article 50 obligations "largely remain on the original schedule" (Gibson Dunn, 2026).

The Commission adopted its final guidelines on Article 50 on 20 July 2026, less than two weeks before the start date (Bird & Bird, 2026). The guidelines are not binding, and only the Court of Justice of the EU can interpret the Act authoritatively. National market surveillance authorities and the AI Office can still be expected to follow them.

How does Article 50 work: who marks and who labels?

Article 50 splits the work in two: the companies that build generative AI mark its output, and the companies that publish that output label it. Paul, Weiss advise every company to map whether it is a provider, a deployer or both (Paul, Weiss, 2026).

1. Providers mark. Under Article 50(2), makers of generative systems must mark outputs so they can be detected as AI-generated. Systems already on the market before 2 August 2026 have until 2 December 2026, and interoperable watermark detection is required by 2 February 2027 (Paul, Weiss, 2026).

2. Deployers label. Under Article 50(4), the brand that publishes deepfakes or qualifying text must disclose it. This duty has applied since 2 August 2026, with no transition period (Bird & Bird, 2026).

The Code of Practice on transparency of AI-generated content shows how to label in practice. The Commission has assessed it as adequate and describes it as the only EU-wide recognised framework for demonstrating compliance with these duties (Bird & Bird, 2026). For published text, the Code points to an icon above or near the headline, or in the colophon. Brands may label only the AI-generated part, and short texts can carry a contextual notice in the interface.

What do brands get wrong about AI disclosure?

The most common mistake is to treat labelling as either a ban or a blanket rule. It is neither.

• "The AI Act limits what we can make with AI." Article 50 requires transparency about AI. It does not restrict what AI can be used for (Paul, Weiss, 2026).

• "Everything AI touched needs a label." The visible duty covers deepfakes and public-interest text, not every asset made with AI.

• "Our tool's watermark covers us." Provider marking and deployer labelling are separate duties on separate parties (Bird & Bird, 2026).

• "We must relabel the archive." Content generated and published before 2 August 2026 does not need to be marked retroactively (Bird & Bird, 2026).

When does labelling apply to our campaigns?

Labelling applies when AI-generated or altered content is used in the EU and either looks real or informs the public on matters of public interest. Use in the EU includes cases where the deployer itself foresees that the output will be spread and used there (Bird & Bird, 2026). A brand based outside the EU that targets EU audiences can therefore be in scope.

Some questions are still open. For evidently artistic, creative, satirical or fictional work, the Act limits the duty to an appropriate disclosure that does not hamper the display or enjoyment of the work (Regulation (EU) 2024/1689, Article 50). Whether a specific campaign qualifies, and when brand copy counts as public-interest text, depends on the guidelines and how authorities apply them. Settle these questions with legal counsel, case by case.

How does IW Agency see AI labels in brand work?

The view in this section is ours, not the law's. At IW Agency, we think the label is a brand touchpoint, and it deserves the same craft as the logo next to it. People choose brands they feel they can trust. A disclosure that looks hidden or bolted on erodes that trust.

We would design the notice once and build it into the brand's design system, with rules for icon, wording and placement in every channel. We also think sameness is a bigger risk with AI imagery than the label. Our guide to keeping AI-generated images and copy distinct covers that side. Our work with Google at Almedalen started from a related wish: to offer a more human perspective on technology as AI shaped public debate.

Conclusion: Design the disclosure like you mean it.

What does this mean for your brand?

• Map your roles. List every AI tool and workflow, and note whether you act as provider, deployer or both.

• Sort assets by trigger. Separate realistic image, video and audio, public-interest text, and everything else. Only the first two groups carry a disclosure duty.

• Design the label once. Set the icon, wording and placement in the design system, in line with the Code of Practice.

• Document human review. If you rely on the editorial-responsibility exemption for text, record who reviewed what.

• Settle grey areas early. Agree with legal counsel how you will treat creative work and public-interest copy before the next brief goes out.

Frequently asked questions

Do AI-generated product images need a label?

They need a label only if they count as deepfakes, meaning they resemble real people, objects, places or events and would falsely appear authentic. A realistic image of a real product in a real-looking setting can come close. Check borderline cases against the Commission's guidelines.

Does the AI Act apply to brands based outside the EU?

It can. Use in the EU includes cases where the deployer foresees that the output will be spread and used in the EU (Bird & Bird, 2026).

Do we need to relabel content published before August 2026?

No. Content generated and published before 2 August 2026 does not need to be marked retroactively (Bird & Bird, 2026).

What happens if we don't label?

Breaches of the transparency rules can lead to fines of up to the greater of EUR 15 million and 3 per cent of worldwide turnover (Paul, Weiss, 2026).

Where does that leave your next campaign?

Labelling is a transparency rule, not a creative limit. The work lies in knowing which assets trigger the duty and in making the label feel like part of the brand. If you want to think it through for your own campaigns, write to us at hello@iwagency.com.

Sources

• Commission publishes guidelines on transparency obligations for providers and deployers of certain AI systems — European Commission, 2026

• European Commission adopts final Guidelines on AI Act Article 50 transparency obligations: first impressions — Bird & Bird, 2026

• Taking the EU AI Act to Practice: The Final Transparency Code of Practice — Bird & Bird, 2026

• EU Finalises Transparency Rules for AI-Generated Content — Paul, Weiss, 2026

• EU AI Act Omnibus Agreement — Postponed High-Risk Deadlines and Other Key Changes — Gibson Dunn, 2026