Should a business label every image, paragraph and brainstorm touched by AI? That would make for a very busy internet.

The first thing to understand is that not every recommendation is currently a legal requirement.

The National AI Centre’s guidance on being clear about AI-generated content is voluntary guidance. It encourages businesses to choose appropriate transparency mechanisms such as visible labels, watermarking and metadata when AI has generated or materially modified digital content.

That still matters.

Voluntary guidance can shape practice and customer expectations. It does not, by itself, create a new legal duty to label every AI-assisted item.

The useful question is therefore not “Do I legally have to label every use of AI?”

It is “Would a reasonable person care that AI created or materially changed this?”

If AI is used internally to brainstorm ten headline options and a human writes the final article, a prominent AI label may add little value.

If a synthetic image depicts a real event that never happened, the case for a prominent disclosure is much stronger. The applicable legal obligations still depend on context.

If an AI system is presented as a human customer-service representative, users may reasonably expect to know.

If AI materially transforms a person’s image, voice or likeness, provenance becomes increasingly important.

Context is everything.

The Government’s guidance describes several mechanisms. A visible label is the simplest. Watermarking can embed evidence directly into content. Metadata can preserve information about provenance without necessarily dominating the visual experience.

For designers, this is not only a compliance consideration. It is an interface problem.

Bad disclosure design can make products unusable. Put a warning badge on every AI-assisted microinteraction and the interface becomes noise. Hide everything in a privacy policy and transparency becomes meaningless.

The better pattern is progressive disclosure.

Give people the information that matters at the point where it matters, then provide deeper details when they need them.

For example, an AI-generated product image might carry a subtle but clear label with an accessible information panel explaining how it was produced. An automated recommendation could provide a short explanation of the role AI played and a path to human review.

Transparency should improve understanding, not merely satisfy a checklist.

This is also where design systems will need to evolve.

We already maintain reusable patterns for errors, consent, cookies, privacy settings and account security. AI transparency will likely become another reusable interaction category: AI labels, provenance indicators, confidence messaging, human-review controls and contestability flows.

Agencies can start testing these patterns now, especially where undisclosed synthetic content could mislead people.

Because the biggest mistake would be waiting for regulation, then bolting a legal sentence onto an interface that was never designed to explain itself.