/

Updated 18 November 2026

How AI Disclosures Affect User Experience (and How to Design Them)

AI Disclosure: This content includes AI-generated text.

Short answer: an AI disclosure costs you a little trust when it’s vague or alarming, and very little when it’s specific and calm. Research on AI labels shows people do rate labelled content somewhat more sceptically, but the effect shrinks when the label explains what the AI did and who checked the result. Since 2 August 2026, some labels are required under Article 50 of the EU AI Act anyway, so the practical question is how to design them well, not whether to have them.

What the research says

Studies of AI labels on news and online content point the same way:

  • A 2024 study by Sacha Altay and Fabrizio Gilardi in PNAS Nexus found that headlines labelled as AI-generated were rated as less accurate and shared less, even when they were true or actually written by humans. The penalty was larger for labels implying the AI did everything.
  • Research from the Reuters Institute and others has found audiences are more comfortable with AI used behind the scenes, such as for translation or editing support, than with AI writing content end to end, and that listing sources and human oversight alongside the label helps.

The lesson for website owners: a blanket “AI-generated” stamp on everything has a cost. A precise label on the content that needs one, with a word on human oversight where it applies, has much less.

When a disclosure helps rather than hurts

  • Chatbots: knowing it’s a bot sets expectations. Visitors phrase questions differently and are less frustrated by limits. Discovering later that “Laura” was a bot does far more damage than saying so up front.
  • Images: a small “AI-generated” badge on a concept visual is barely noticed. A product photo later revealed as AI erodes trust in the whole shop.
  • Articles: “Drafted with AI, reviewed by [name], tax adviser” can read as a sign of a careful process.

Design patterns that work

1. Be specific

“This article was drafted with AI and fact-checked by our editorial team on 12 November 2026” beats “AI content”. Specific labels answer the reader’s real question: can I rely on this?

2. Keep it calm

Use your normal text colour or a subtle tinted box with a small icon. Red warning colours and exclamation marks suggest a problem you’re confessing.

3. Put it where it’s needed, and only there

Label the specific content that requires it: the chatbot, the photoreal image, the unreviewed public-interest article. A site-wide banner saying “we use AI” adds noise without informing anyone. See where to place AI disclosures.

4. Show it at first exposure

Article 50(5) requires disclosure no later than the first interaction or exposure. From a UX view, early is also better: a label found at the end of an article feels like small print.

5. Make it accessible

Real text, sufficient contrast, readable by screen readers, and not hover-only. That’s also a legal requirement under Article 50(5).

6. Use familiar symbols

The EU has published common icons for labelling AI-generated content, and the Code of Practice recommends them. As they become familiar, recognition will improve. Always pair the icon with words for now.

Examples

Context Weak Better
Chatbot “Hi, I’m Laura!” “Hi, I’m Laura, the AI assistant for Acme. Ask me about orders, or type ‘human’ for our team.”
Article “Content may be AI-generated.” “Written with AI assistance. Not reviewed by an editor.”
Reviewed article (no label required) Nothing “Reviewed by Jana Weber, data protection lawyer, 12 Nov 2026.”
Image Disclaimer in the footer Small “AI-generated image” badge on the image and in the caption

Measuring the effect on your site

If you’re worried about engagement, measure rather than guess. Compare time on page, scroll depth and conversions for labelled pages before and after the change, or label a sample of posts first. Heatmap tools show whether visitors notice the label at all. In most cases the difference is small next to the effect of content quality itself.

What not to do

  • Don’t hide required labels to protect metrics. It’s a legal risk and a bigger trust risk.
  • Don’t label everything “just in case”. Over-labelling dilutes the signal and invites the scepticism research describes.
  • Don’t use deceptive personas. A named bot is fine; a bot that claims to be human isn’t.

The free EU AI Act Ready plugin for WordPress shows a notice only on posts whose disclosure level needs one, lets you choose a banner or a subtler style, before or after the content, and lets you write your own wording. Our Article 50 content test helps you decide which posts need a label in the first place.

Frequently asked questions

Will an AI label hurt my SEO?

No. Google has said it rewards helpful content regardless of how it’s produced, and a disclosure label isn’t a ranking signal against you. Thin or unhelpful AI content is the problem, label or not.

Can visitors close the label?

For voluntary labels, yes. For a legally required one, keep at least a small persistent marker so the information stays available.

Should I label AI-assisted content even when it’s not required?

That’s a trust decision, not a legal one. If you do, keep it specific about what the AI did. See the AI transparency rules for websites for what’s actually required.