From mellotrons to marketing: The AI labelling debate

A couple of years back, I was getting more comfortable with the idea that if an art director uses AI to help generate a background, an animal or a generic prop, or helps create a stunt, why should that be any different from using stock photography, Photoshop or CGI? No label necessary?  

Then I was asked to advise on a storyboard featuring a seemingly human protagonist who wasn't a real person at all.  Something felt off.  By that stage, we were already moving beyond the obvious AI giveaways: the “too many fingers to be real” clues were disappearing.  I pondered on the use of undisclosed synthetic humans in ads. Does it matter? If so, when does it matter?   

When deciding whether a disclosure or label is needed for legal or regulatory reasons, for me,

the question shouldn’t be whether an AI tool has been used or not.  The question should be, what difference does the AI make to the audience's understanding of the message?

The challenge is preserving the value of disclosure by reserving it for situations where consumers genuinely need the information.  Otherwise, they will start tuning it out in a sea of noise.  Real risks will be missed.

The direction of travel in Europe has been towards mandatory disclosure. The definition of “deepfake” in the EU AI Act is much broader than what English speakers may naturally interpret as “deepfake”.  A deep fake is defined in the EU AI Act as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful.  This captures a much wider range of AI-generated or AI-modified content than the term "deepfake" commonly implies in the UK, where it is often associated with malicious uses such as impersonation or deception.

It seemed to me that the initial assumption in the EU when drafting the law may have been that content is either fully AI or fully real.  However, many campaigns are hybrid and combine both real and AI generated elements.  The most recent EU guidance does include some helpful advertising examples now but it doesn’t address many of the more nuanced uses of AI humans in advertising.  Given the breadth of the deepfake definition and significant fines, advertisers and agencies may need to take a cautious approach and assume that most AI talent use in the EU requires disclosure unless the guidance clearly indicates otherwise. I consider that in some circumstances, we end up in a situation where AI labelling and disclosure in the EU has the potential to actually confuse or even mislead the consumer rather than assist understanding.

People like to bandy around phrases such as “it’s a wild west” and “it’s an unregulated space”.  Pause.  Breathe. And let’s take a look at the current CAP and BCAP advertising codes.   

At the time of writing, the UK does not have an equivalent of the EU AI Act, although AI labelling remains under consideration by government. However, before we declare AI advertising a regulatory free for all here, it's worth looking at the rules we already have.

We already have an established and responsible advertising self-regulatory system in place in the UK and many of the tools needed to tackle harmful uses of AI already exist in the CAP and BCAP Codes. The essentials are already there. In fact, the more I think about it, the more I find myself reaching for familiar advertising principles rather than wishing for an entire set of new AI-specific rules.

If an AI-generated ad creates a misleading overall impression, the existing rules on misleading advertising and material omissions are likely to be relevant. In some cases, the use of a disclosure or label may not be enough to solve the problem. A disclosure can inform and clarify, but it cannot simply contradict the overall impression created by the ad.

If the use of AI raises broader concerns about harm to consumers or society, the social responsibility rules may come into play. If AI is used to exaggerate what a product can do, advertisers face the same substantiation challenges they have always faced. And if an ad relies on what appears to be a genuine testimonial, endorsement or personal experience, there are already rules requiring such claims to be genuine and not misleading.

Importantly, many of these advertising rules are underpinned by consumer protection legislation. That means it is not just the ASA that may take an interest. The CMA has already shown a willingness to intervene where consumers are misled, including in relation to hidden influencer marketing and fake or incentivised reviews.

You be the Judge

You be the Judge: My thoughts

Consumers don't need to know every time AI is used. They need to know when it matters.

Not all AI humans are equal.  Some AI humans are used as production techniques.  Other AI humans are potentially deceptive because the audience may believe they are hearing genuine, lived experience or because the audience believes they are forming a relationship with a real person.  I can see the potential for harm with undisclosed AI influencers in the spheres of health and finance and when the audience is a vulnerable group.  

For me, for the UK, I consider that, 1, and 2 are a no label necessary situation.  3 and 4 either need disclosure or may still mislead even with a label.   5 and 6 are interesting: viewpoints will vary. There is an argument that there is no real consumer harm here and therefore no need for a regulatory disclosure.  However, advertisers may still choose to disclose for ethical, PR or brand trust reasons.

Do we succumb to blanket, labelling that treats huge rafts of AI creativity like a hazard? Or will we push for a smarter standard based on real harm?

Let’s work out a sensible approach to AI labelling to allow creativity to thrive without any onslaught of unnecessary visual clutter, which will soon lose its intended impact if not employed with care.  Let’s concentrate on disclosures where there is potential for real harm.   

The wholesale adoption of the EU labelling approach is not the best answer for advertising in my book. The Blue Peter “one we prepared earlier” from the EU should not be a shoo-in for the UK government.  We can be a bit more Moog.   

IPA Legal and AI

Every day, the IPA Legal Team responds to IPA member requests for advice on advertising and employment law issues. The team also devotes considerable time to keeping members informed of developments affecting marketing communications, employment law and, where relevant, the wider business environment.

Guidance on generative AI is a key area of focus, covering contract provisions, copyright, deepfakes, synthetic humans, governance and the application of industry principles. Team members have also contributed to a number of industry AI taskforces, including engagement with the Government AI & Labelling Workstream on transparency, disclosure and labelling requirements, and their potential implications for advertisers, agencies and the wider creative industries.

IPA legal support