What the EU AI Act means for AI-Generated Product Content

Ocula CTO and Co-Founder Greg Fletcher breaks down how the EU AI Act affects Ecommerce brands publishing AI product content, when to take action, and why humans and AI should be working together as part of your review process. 


If a shopper asked whether your product content was written by AI, would you tell them?

If you work in ecommerce and your content reaches people in the European Union, you’ll want to read this blog. I'll break down when AI-generated product content needs to be disclosed under the EU AI Act, and why humans and AI should be working together as part of your review process. 

What does the EU AI Act mean for ecommerce brands?

Recently, there’s been a lot of noise about the EU AI Act, the world’s first comprehensive AI Law. But what does it mean for Ecommerce brands publishing AI content online? 

Article 50 of the Law sets out clear AI transparency rules for brands publishing AI content. It applies to anyone whose content reaches people in the EU, even if you’re based elsewhere. 

Who is responsible for marking AI-generated content?

When creating AI product content, most of the responsibility to disclose AI’s role within that process lies with providers. Providers are the teams that build the agents that generate the AI content. These are teams like mine, at Ocula.

Providers must mark AI outputs as AI-generated (sometimes called AI watermarking). At Ocula we’re doing that by adding machine-readable metadata on every piece of content we produce, covering version, timestamp and source. Providers whose tools were already on the market before August have until the 2nd of December 2026 to get this live.

Do retailers need to label AI-generated product descriptions?

For most retailers – or anyone publishing AI content, legally known as the “deployer” – you don’t need to label your product content (whether product descriptions, metadata, titles or attributes) as AI-generated under the EU AI Act. 

The exception is for product content that touches on areas of “public interest”, like health and safety instructions or environmental claims. 

In these cases, AI product content may need to be disclosed –  unless your brand takes editorial responsibility for what goes live, with a human editor who deliberately reviews your content and has the authority to change it. The European Commission calls this human review or editorial control.

So even for your most sensitive products, a properly empowered human editor means there’s no need to label your AI content under the Act. 

If you sell thousands of products in the public interest category, but have a small team and don’t have a human overseeing the process at any point, some compliance vendors recommend you blanket label all product content as AI generated.

Bear in mind that this is a precaution rather than a strict legal reading. And it comes at a price: blanket labelling can harm brand voice, perceived value, and erode consumer trust. 

Lessons we can learn from GDPR

Remember the early days of GDPR – another privacy and security law created by the European Union? 

In 2018, when it came into force, every brand you'd ever bought from was suddenly in your email inbox, asking you to click a button if you wanted to continue to receive marketing emails.

In reality, a lot of those emails did not need to be sent. Many of these businesses already had a perfectly good lawful basis to keep emailing their customers, but, of course, the customers didn’t click the marketing button. 

And so brands ended up deleting large segments of their email list for no legal gain at all. 

Blanket labelling AI content risks doing the same thing. Overcorrecting doesn't always make you safer. It can make your content slower to produce, lose segments of your audience, and give your team more work. 

How to use AI governance agents to keep product content compliant

The better option is going beyond AI content generation, and using your AI governance agents to automatically flag anything risky in your product content that needs a review. 

Set up your agents to flag any product that touches on health and safety, makes an environmental claim, or breaks your brand, tone or enrichment sourcing rules. 

Make sure these Governance agents escalate this content straight to your human editor for review. All other content can go live as is, with your editor spot-checking a sample each week. And every flag shows you where your AI rules need tightening.

This way, your editor only reviews what matters, you stay on the right side of the EU AI Act, and you keep on top of product content in the race to win on AI Search, from Google AI Overviews to ChatGPT. 

And that’s how humans and agents can work together to scale AI product content without the risk.


Read more like this

Next
Next

How The Sole Supplier made 45,000 products discoverable for shoppers, search engines, and AI