DO NOT DELETE THIS LOADS THE ROCKSTAR FONT
Article 50 of the EU AI act became applicable on the 2nd of August 2026 – why? To stop people from being deceived about what is real and what never happened.
AI has a use across almost all businesses and it’s becoming increasingly adopted.
Marketing and communications feels like an area that’s adopted this new tool heavily and quickly. With good reason too, there’s a lot to marketing that’s process-heavy, resource-demanding, time-consuming and distracting from being truly creative. However, everyone in marketing and communications has a responsibility to do what they do in an ethical, morally-acceptable, true, honest, fair and of course, legally compliant way.
Within a few minutes of being on any social media platform, you’ll likely see an ad or a piece of content that you’d consider to be AI slop. In this article, we’re not talking about the fake talking cat videos, the impossible DIY videos, Will Smith eating spaghetti or the historically-inaccurate “what life was like as a Roman” videos. We’re talking about the content being produced and used by marketing teams.
The transparency rules in Article 50 of the EU AI applies to two categories:
| Provider | Builds an AI system and puts it on the market. Your vendor. Their job is to build in disclosure and mark outputs in a machine-readable format. |
| Deployer | Uses an AI system professionally. You. Your job is to visibly disclose content that could be taken as real, and to label certain AI-written public-interest text. |
The majority of people reading this would fit into the deployer category. Providers are vendors of AI systems. For example, if you license an AI chat for your website, the company you’re buying it from is the provider and you are the deployer. If you have built your own AI chat bot to use on your website, you’re both the provider and the deployer.
So, as a deployer, you now need to label the AI content you’re putting out into the world. There are various levels to this and various scenarios where you would not need to label content.
The intent of the content you’re producing doesn’t matter, if you have created a visual asset that depicts a real person, thing, environment, event, venue, crowd, etc – these fall into the deepfake category. Deepfake is quite a heavy-sounding word, and has very negative connotations – and it’s understandable why, deepfake videos have caused real harm; from pushing damaging far-right narratives, spreading fake rumours about public figures and more. Unfortunately for us marketers using AI content generation tools to do unharmful things like create “stock” imagery to use in ads, be able to create photos of people shaking hands that actually live on opposite sides of the world and place products into an environment to illustrate how it’s used, these are all considered deepfakes and must be labelled correctly.
The question is “would someone think this is real?”. Or is it clearly not mistakable as authentic, like product renders. Pre-AI you can tell they’re not real photos of products, they’re renders based on CAD files – if you do this now using AI, the outcome is the same – they still don’t look like real photos of the product, so you’re most likely in the clear. If you’re risk-averse, label it.
The EU has published an official icon set which they say perform better when paired with a short piece of text. They are:
| Icon | When to use it |
 | AI was involved in creating the content, and you are pairing the icon with your own text label or a second interactive layer. |
 | The whole thing was made by AI, with no human-created elements and no editorial control beyond the prompt. |
 | Existing human-made content that AI altered, turning it into something that could be taken as real. |
https://digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content
Placement matters too – there’s little point using something that is meant to demonstrate transparency and then hiding it away somewhere. Icons should be:
For accompanying text, short and unambiguous beats careful and hedged. AI-generated. AI-modified. Voice generated with AI. Visualisation, AI-generated, not a photograph of an installed system. Created with the help of AI is weaker, because it invites an argument about how much help.
The key thing here is whether someone could reasonably believe the image is real. Abstract backgrounds, artistic-style illustrations, things evidently not real – like a dragon flying over a stadium, do not need to be labelled. Only if the image could be real. Placing your product on the moon, turning it into a hot air balloon, or putting your bosses face on Godzilla, do not count.
It’s not just visual content either – copy is also covered by these new rules. However, there’s an important distinction; it only applies to copy that’s considered public interest. Which is narrower than it sounds – it’s largely aimed at automated news. Topics around politics, public health, environment, public security and similar things are what’s being targeted here. Not marketing copy. And even in the case of public interest, if an individual person is taking editorial responsibility and has actually checked it for factual correctness, not just grammar, then it’s had a thorough human review and does not need labelling.
| Likely needs a label | Likely does not |
| Adding kit to a half-empty rack | Background removal, retouching, colour grading, upscaling |
| Dropping speakers or screens into a real venue photograph | Changing the colour of a shirt in a real photograph |
| A photorealistic render of an install shown as a completed project | A product visual generated from CAD that shows the product accurately |
| Putting a product into someone’s hands that was not there | Evidently illustrative concept imagery |
| An AI-generated audience or crowd shown as documentation of an event | AI used in early-stage ideation that never ships |
| A synthetic voice presented as a named person or customer | Copy drafted with AI and then genuinely reviewed and signed off |
Short answer: yes.
It doesn’t matter where you’re based or where the content was made. What matters is whether the output is used in the EU. If you’re selling into Europe, exhibiting at European shows, working through EU distributors or running campaigns that deliver to EU audiences, you’re in scope. A campaign that’s geo-restricted to a region outside the EU is, in theory, outside it.
If you’re out to deceive people with fake imagery generated by AI, demonstrating a product doing or performing in a way that is untrue, then tough luck – you were likely already breaking laws but now you’re also breaking this one.
For small marketing teams where budgets are tight, and for niche markets where stock imagery is often limited and application imagery is often lacking in volume, this feels like a bit of a knock. With thorough human review, good prompting and consistency, it’s possible to create good marketing imagery at a fraction of the cost of organising a photo shoot. It doesn’t mean you can’t do this, but it does mean you need to label it.
Regulation is a word that’s being thrown around a lot in the world at the moment as something that’s negative – but we need it. Fair regulation is a good thing.
Commission guidelines on Article 50 transparency obligations
Commission FAQ, plain-English answers on scope and definitions
EU icons for labelling AI-generated content, including downloads
Code of Practice on Transparency of AI-generated Content
This is not legal advice – if you are unsure on anything, seek proper legal advice. We’re happy to talk marketing and AI, but we’re not lawyers (thankfully).
It’s not about intent. It’s about whether someone could believe it’s real.
