Resources · Guide

Does my business need to make AI disclosures?

A practical way to work out whether Article 50 is likely to affect your business, your website, and the content you publish.

Take the quiz

Not every business using AI needs to disclose every piece of content. The rules depend on what you publish, how AI was used, and who can access it. Answer a few quick questions and we'll assess whether the EU AI Act is likely to apply to your business.

AI Disclosure Assessment

Does your AI content need a disclosure?

Answer a few questions about how you use AI and we'll explain whether the EU AI Act's transparency requirements are likely to apply.

Takes around 2 minutes

Step 1 of 4

Who is your content intended for?

This helps determine whether the EU AI Act transparency rules may apply.

Choose one

Four questions that determine whether disclosure is likely

Although every situation is different, most businesses can begin by asking four practical questions.

1. Do you reach people in the EU?

If EU customers, readers or users are part of your audience, the Act can apply even if your business is based in the UK or US.

2. Are you using AI to generate or materially alter content?

The duties focus on outputs people see or hear, not on private internal brainstorming or light back-office assistance on its own.

3. Could the result look or sound authentic?

For image, video and audio, the main issue is whether an ordinary person could take it as real rather than clearly stylised, fictional or trivial editing.

4. Is the text informing the public on a matter of public interest?

Text has a narrower rule. Marketing copy usually falls outside it; public-interest publications without real human editorial control are the harder cases.

Who is most likely to be in scope?

In practice, the businesses most exposed are the ones publishing outward-facing content: ecommerce brands, agencies, marketplaces, publishers, media teams, creators, app companies and service businesses using AI visuals, AI voice, AI video or automated public-facing text.

Common commercial examples

- AI product photos, model shots or lifestyle images
- Real photos with AI-added objects, people or backgrounds
- AI voiceovers, cloned voices or realistic audio ads
- Explainer videos or demos using realistic synthetic footage

Public-information examples

- A public weather warning generated by AI
- A published article or bulletin about health, safety or elections
- A public crisis update written or materially rewritten by AI
- A newsroom-style output with no real editorial sign-off

When the answer is usually yes

Likely disclosure territory

- You publish realistic AI-generated or AI-modified images, video or audio
- The content is aimed at, accessible to, or used with people in the EU
- The AI change is substantial rather than routine editing
- The text is published to inform the public on matters of public interest
- No meaningful human review or editorial control happens before publication

Often outside the rule

- You only use AI for internal ideation or drafting and do not publish the output as-is
- The content is clearly fantastical, satirical or otherwise not plausibly authentic
- The AI use is standard assistive editing that does not substantially change meaning
- The text is ordinary product copy, entertainment or a private one-to-one output
- The publication underwent genuine human review or editorial control with responsibility

The special rule for text

Text is where many businesses either over-worry or under-worry. Article 50 does not say every AI-written paragraph needs a label. The narrower text rule applies when AI-generated or AI-manipulated text is published, intended to inform the public, and concerns a matter of public interest.

Draft guidance gives examples such as public safety warnings or similar public-interest information. It also points the other way: fictional novels, private outputs, and ordinary advertising copy are not the main target. If a real human editor can approve, alter or reject the substance before publication, and someone holds editorial responsibility, the disclosure duty can fall away for text.

A useful practical test

Ask: are we publishing this to inform the public about something that matters beyond our own promotion, and is a real person substantively accountable for what goes live? If the answer is no, pause and review it.

Common misconceptions

Mistake: 'We are not in the EU, so it does not apply.'

The scope question is driven by where the affected people are. A non-EU business with EU customers can still be caught.

Mistake: 'We used AI once, so everything needs a label.'

No. The rules are targeted. The key issue is the nature of the output and whether the artificial origin should be disclosed.

Mistake: 'Our editor skimmed it, so the text exception is covered.'

The guidance points to substantive human review or editorial control, not a token glance after the fact.

Mistake: 'If we label it, every other legal issue disappears.'

Labelling helps with transparency, but it does not resolve consent, copyright, defamation, consumer law or platform-rule issues.

What to do if you might be in scope

If your organisation publishes AI-generated or AI-modified content, the safest approach is to:

Identify where AI is being used.
Review which content may fall within Article 50.
Decide where disclosures are appropriate.
Apply disclosures consistently.
Keep a record of how those decisions were made.

The Commission's guidance repeatedly emphasises transparency, consistency and helping people understand when they are interacting with AI or viewing AI-generated content.