The degree of labelling necessary, if any, is going to depend on where you are and where the consumers who view the ad are. We’ve looked at the position in the UK and how UK advertisers may still be impacted by transparency requirements under the EU AI Act.

UK – to label or not to label

The UK does not have any artificial intelligence (AI) specific regulation. Uses of AI are therefore governed by existing laws, depending on the use.  

The relevant advertising law is the prohibition on unfair commercial practices in the Digital Markets, Competition and Consumers Act 2024, which is also reflected in the Advertising Codes enforced by the Advertising Standards Authority. Unfair commercial practices include misleading consumers by action or omission. 

So while there is no explicit reference to the use of AI or labelling AI-generated content in the law or the Ad Codes, advertisers need to consider whether consumers will be misled if the use of AI is not disclosed.

In a nutshell this means that AI cannot be used to:

• Provide false or misleading information relating to a product/service or the effect of that product/service.
• Create an overall presentation which is likely to deceive the average consumer about a product/service or a transactional decision.
• Omit information that the average consumer needs to take an informed transactional decision.

It’s vital to remember that 'transactional decision' is much wider than a decision to purchase and can include deciding to visit a shop or click through on a website, and that 'omission' includes providing information in an untimely or unclear way.

Crucially, labelling AI use will only assist advertisers where the use does not cause consumers to be misled. If an AI-generated image of a product includes functions the product doesn’t have in real life, then labelling it as ‘AI-generated’ will not stop the image from being misleading in terms of functionality. The ASA ruled against an ad which highlighted how realistic a toy was, (‘it's like having a real puppy’), but used AI to generate visuals of the product. The AI version of the toy depicted did not reflect the functionality and appearance of the toy on sale and the ad was found to be misleading.

As with any qualifying information, labelling that AI has been used needs to provide clarification rather than contradict the claims.

There are circumstances where using AI doesn’t need to be labelled, for example if it isn’t material to the ad. The principles which apply to traditional post-production techniques are likely to be cross applicable. The ASA will assess advertising on a case-by-case basis and compliance will always depend on the final execution.

EU – transparency requirements 

In the EU, the new requirements for transparency under the EU AI Act include rules in relation to "deep fakes" with a wider reach than one might expect. It’s not just replications of real people. A deep fake can be a wholly fictional AI-generated model in a shampoo advert, a distant place, or a product shot that was never actually photographed but AI-generated. No real, identifiable person needs to appear in the image. 

If an ad is caught by the transparency requirements, the fact that the content has been generated or manipulated by AI needs to be disclosed.

The deep fake rule sits in Article 50(4) of the AI Act, one of a small set of transparency obligations that applied from 2 August 2026. European Commission Guidelines show how it considers those obligations should work in practice and are intended for use by regulators as well as businesses – please see more below. 

Brands can become signatories to the Code of Practice on marking and labelling of AI-generated content to help demonstrate their commitment to compliance with the EU AI Act obligations. 

Why the EU AI Act matters even if you are UK-based

The EU AI Act doesn’t apply in the UK, but UK advertising could be caught. 

The Act talks about content which can be "accessed in the EU". So advertising from UK companies targeting EU consumers could be caught, even if there is no EU entity. Even content on the open internet, where EU reach is a foreseeable and realistic possibility, might be enough on its own, unless the UK can show EU exposure occurred through channels it neither foresaw nor controlled. 

(UK advertisers who can be confident that AI-generated content will not be accessed in the EU can stop reading here).

What does the EU count as a deep fake?

In order for image, audio, or video content to be considered deep fake, four conditions need to be fulfilled. The content:

1. resembles; 
2. existing; 
3. persons, objects, places, entities or events; and 
4. it would falsely appear to a person to be authentic or truthful. 

The transparency Guidelines go through each of them in detail. From an advertising perspective, two limbs require a closer look:

• 'Existing', doesn’t only mean 'real': It is enough that the content resembles something that exists, could plausibly exist, or could plausibly have existed. A wholly fictional, entirely made-up AI-generated person can meet this test just as easily as a digital replica of someone real. From an advertising perspective, the broad definition of ‘existing’ could push an AI-generated ad into the scope of the deep fake definition. 
• The Guidelines treat falsely appearing as authentic or truthful as two different tests: 'Authentic' asks whether the content genuinely is what it claims to be, in how it was actually made. 'Truthful' asks whether what is depicted is factually accurate. The European Commission's own guidance draws quite a practical line here for product advertising: ordinary AI polish, such as colour correction or a tidied-up background, has only a minor effect on how authentic something looks and is not treated as a deep fake. An AI-generated product image that makes the product look better, different, or of higher quality than it actually is more likely to be a classed as a ‘deep fake’. It’s easy to foresee challenges stemming from where to draw this line. 

The wording of the transparency Guidelines indicates that it all comes down to assessing the whole piece of content to determine whether it should be considered to be deep fake or not. 

The creative exemption is not guaranteed

There’s a tempting-looking exception which allows for lighter-touch disclosure where content is evidently artistic, creative, satirical or fictional. However, the Guidelines say that content which is primarily commercial in character generally does not benefit from this exemption, so this is unlikely to help advertisers.

Text can trigger the requirements too

Although the focus is mainly on visual media, there are circumstances where text is also caught. Text which has been: 

• generated or manipulated using AI;
• is published to inform the public on matters of ‘public interest’; and
• has not undergone human review/editorial control nor is subject to editorial responsibility. 

The Guidelines treat a wide variety of issues as being in the public interest which means that advertising copy referring to health, consumer safety or sustainability which was generated or manipulated by AI may need to meet the disclosure requirements.

It could be quite easy to unintentionally cross the line here. 

What is adequate disclosure?

Labels must be clear, distinguishable and compliant with accessibility requirements. The law itself does not mandate a specific label, however, the EU has produced icons for labelling AI-generated content. These are designed to support meeting the requirements of Code of Practice on marking and labelling of AI-generated content. 

The icons include:

• AI basic icon: When AI was involved in the creation of deep fake content (image, audio, video) or published text or when a custom text label or interactive second layer is implemented.
• Fully AI-generated: When the entire deep fake content (image, audio, video) or text is fully generated by AI with no human-created elements or human editorial control (apart from prompting).
• Partially AI-modified: When pre-existing, human-made content was partially modified with AI turning it into a deep fake or text on matters of public interest.

View the icons on the EU website

When should disclosure be made – early, and sometimes more than once

The duty applies at the latest at first exposure to the AI material. 

This is particularly important for video and social content. Where viewers may arrive mid-clip rather than at the start. A single disclosure at the beginning might not be enough on its own if it is foreseeable that some of the audience will miss it. Where that is a realistic risk, which for most social content it is, disclosure needs to be repeated or placed so that it is seen regardless of when someone actually starts watching.

The judgement calls are the advertiser’s to make, and to explain

Article 50(4) does not give advertisers a checklist to work through. It hands them at least four separate judgement calls, and each one sits squarely with the advertiser to make and then explain to the regulator if challenged. Four questions to always be ready to answer are: 

1. Does the content resemble an existing (or potentially existing) person, object, place, entity or event? 
2. Would it appear authentic or truthful to the audience actually likely to see it?
3. If the exemption for satirical and creative works cannot be applied? 
4. Has the correct disclosure been applied at the right time?

It’s worth remembering the EU AI Act applies to a wider scope of activities beyond advertising. Watch this space for more insights from the data protection team on the bigger picture for AI disclosure.