We've spent the last few years sticking a little sparkle icon on AI features and calling it transparency. A shimmer here, a tooltip there — done. But here's the thing: that was never transparency. It was decoration. And regulators have noticed.

So what's actually changed?

The EU has issued new guidelines on how AI-powered products and features need to be labelled. The upshot: vague, decorative indicators won't satisfy the requirements. Labels need to be explicit, contextual, and genuinely informative. Users must clearly understand when they're interacting with — or consuming — AI-generated content.

The full implementation details are still emerging, but the direction is unmistakable. If you're building products for a European audience, and most of us are, this affects your design decisions now.

This isn't a legal problem — it's a design problem

Here's what makes this interesting: the guidelines aren't prescribing specific UI patterns. They're setting an outcome — meaningful user understanding — and leaving it to product teams to figure out how to get there. That makes it a design challenge, not a compliance checkbox.

And it's a challenge we should have been tackling already.

Think about how most products currently handle AI disclosure. There's a sparkle icon next to a feature name. Maybe an "AI-powered" badge somewhere in the marketing copy. Occasionally a brief tooltip if you hover in exactly the right spot. But does the user actually understand what the AI is doing? Do they know whether the text they're reading was written by a person or generated by a model? Do they grasp the limitations?

Usually, no. Not even close.

We've been treating AI labelling as a branding exercise — a way to signal innovation rather than inform users. The sparkle icon doesn't say "this content was generated by a language model and may contain inaccuracies." It says "✨ ooh, fancy tech ✨." Those are very different messages.

What does honest AI labelling actually look like?

It depends entirely on context. That's what makes this hard — and what makes it real design work rather than a one-size-fits-all badge.

A search result that's been AI-summarised needs different disclosure than a product photo that's been AI-enhanced. A chatbot response needs different labelling than an AI-suggested email subject line. The stakes are different, the user's expectations are different, and the consequences of misunderstanding are different.

Some questions worth asking about every AI-powered feature you're designing:

  • What specifically is the AI doing? Generating content? Ranking results? Making a recommendation? Each deserves different communication.
  • How reliable is the output? If the system is guessing, the user deserves to know.
  • If not, why not?