Label it or risk it: what the EU’s AI rules mean for UK marketers
Since 2 August, marketers using AI to make content that reaches EU audiences have had new rules to follow. Article 50 of the EU AI Act, covering transparency, took effect on 2 August as planned, even though the Digital Omnibus pushed back other parts of the Act.
What marketers will meet first is the labelling duty deepfake labelling and disclosure when people interact directly with AI apply immediately.
It matters in the UK because the rules follow the audience, not the company’s address. If a campaign can reach EU users, EU reach is foreseeable. The UK approach is different in kind. The ASA is principles-based and asks whether leaving out disclosure would mislead, whereas the AI Act is prescriptive. A brand can be fine at home and still exposed in Europe.
It also matches something audiences already expect. In our recent research (around AI usage in marketing, which will soon be released) clear labelling was one of the things respondents said would make them trust a brand’s use of AI. The regulation is catching up with them.
What the rule asks of marketers
Article 50 covers four situations: AI interacting directly with people, AI-generated synthetic content, emotion recognition, and deepfakes or AI text on matters of public interest. For marketing teams, the last one is the main concern. The deepfake labelling obligation applies to advertising, and the Commission reads the term “deepfake” broadly. Article 50 sits in the second tier of the Act’s penalty framework.
Where it applies
Online video, including YouTube and brand websites: most exposed. EU users can access it directly, so EU reach is foreseeable.
Social media (Instagram, TikTok and similar): in scope. Platform labels may not be enough, especially if they sit only in invisible metadata, so build the label into the ad itself.
Display and other online placements: hard to keep within one market, so treat as in scope.
TV and radio broadcast: the one area where geography can work in your favour, since it can often be confined to one market. That rarely helps once the same campaign runs online.
Audio-only ads (radio, podcasts, smart speakers): need a short, plain-language spoken disclaimer at the start, repeated after ad breaks.
What needs labelling
AI-generated or manipulated images, audio and video that look authentic to an ordinary viewer, whether or not they depict something that really exists
Synthetic spokespeople, influencers and avatars, such as an AI celebrity or an AI version of a CEO
AI-generated product imagery that makes a product look better, different or more capable than it is
Realistic AI-created people, characters, products or audiences presented as real
AI-generated voices and music that could pass as real
AI text on matters of public interest, such as health, safety or environmental claims (not routine copy)
What doesn’t
Cosmetic edits: colour correction, sharpening, re-scaling, minor background extension or tidying
A real product against an AI-generated background, provided the product isn’t misrepresented
Clearly fantastical content, such as talking animals, that no reasonable viewer would take as real
Routine ad copy, headlines and search ads, unless they make a substantive public-interest claim, such as a specific health or sustainability claim
A lighter regime
Evidently artistic, fictional or satirical work, such as film trailers or clearly satirical imagery, still needs disclosure, but in a lighter form. The exemption is narrow, and commercial or persuasive ads, such as teleshopping-style endorsements, don’t qualify.
Timing
Image, audio and video: content created before 2 August 2026 doesn’t need to be labelled retroactively.
Public-interest text: the trigger is the publication date, so text published on or after 2 August needs labelling even if it was written earlier.
What to do now
Audit the AI-generated assets in any campaign that can reach EU audiences
Build labels into the creative itself rather than relying on platform settings
Put your labelling rules in your public AI policy, so one rule covers every channel
Keep a record of what was labelled and why
If you’re doing those already, you’re ahead of the regulation. If not, our AI Readiness Checklist & Whitepaper (soon to be released) is a good place to start.