Key Takeaways:
- In the UK, the ASA banned “recycled” ads from Adidas, Calvin Klein and Uniqlo, with the rationale that their claims were not substantiated by evidence. Elsewhere, EU Commission testing found nearly 40% of labels misstate their fibre mix, casting doubt on the efficacy of labelling in a well-worn discipline.
- Across a more experimental and potentially difficult frontier for labelling – a month from today the EU will enforce AI-content labelling under Article 50, and New York’s “synthetic performer” law goes further still. Governance and disclosure seem inevitable, but the letter of the EU rules may still exempt fully-general models in Europe, leaving fashion’s fastest-growing use case relatively unchecked.
- The AI Act also lets human-reviewed written content skip labelling, leading to a potential short-term scenario where accountability is offloaded to the creatives who sign off on content, rather than the brands themselves.
Summarise and debate with AI:
Take the content and context of this article into a new, private debate with your AI chatbot of choice, as a prompt for your own thinking. (Requires an active account for ChatGPT or Claude. The Interline has no visibility into your conversations. AI can make mistakes.)
This week fashion received another in a long series of reminders that outward-facing claims and verifiable evidence are not the same thing. And while this one came in a relatively unremarkable domain (advertising of products containing recycled materials), the very near future will be defined by similar alarms sounding across a vast and fast-evolving new frontier of content authentication, synthetic media, and AI.
In sustainability, specifically, the existence of a gap between what brands are comfortable committing to advertising copy and what they can actually prove is hardly news. That gap is where the term “greenwashing” comes from, and after a lengthy period of being implicitly permitted to market collections and products to consumers using loose language like “climate friendly” and “green,” the UK’s Competition and Markets Authority (CMA) introduced the “green claims” code, in 2021, to put strictures in place around the way companies can speak about product-level sustainability metrics.
The accompanying guidance to that code contains an evergreen description of the distance between a statement and a data-backed disclosure, referring to false or misleading environmental claims as giving “the impression [products or services] are less harmful or more beneficial to the environment than they really are”.
That definition has been tested numerous times in the last five years, and although enforcement is still patchy and conducted on a review-by-review, complaint-by-complaint basis, the effect has been to at least partially chill fashion’s willingness to make claims about the provenance, impact, or general sustainability profile of products. A good amount of brands, today, choose to sidestep the whole language quagmire and only publicise data when it originates with partnerships with inspectors, auditors, or intermediaries, allowing them to put a recognisable badge in place of a self-attested statement.
But as we saw in the last week or so, that chilling effect has not been universal, and advertising regulators are further tightening their hold on how sustainability is (or isn’t) understood by the buying public.
Before you read on...
Our weekly news analysis will always be available to read here at The Interline, but you can get it (along with notifications for new podcast episodes, events, and more) in your inbox by signing up to our mailing list.
At the end of June, the UK’s Advertising Standards Authority (ASA) ordered a brace of brands – including Adidas, Calvin Klein, and Uniqlo – to take down advertisements that included claims that some of the products featured or alluded to were “recycled”. Based on the ASA’s own language, the root cause of the complaint is that the term “recycled” has an absolute meaning that is poorly-defined in product and advertising; people viewing the ad, the regulator says, could reasonably come away from that exposure thinking that they were purchasing products that were wholly made of recycled material
In their responses, the brands cited certifications (again: shifting the emphasis to a third party) and made claims that the products in question were made to a “meaningful extent” from recycled materials.
Despite earning some headlines, this issue seems easy to solve. Brands can tweak the offending ads to either hedge their wording more carefully, or they can drop the claims and accept that they may cede a small amount of ground to competitors… at least until it emerges that those competitors are under the same microscope.
But taken as part of a body of evidence that labelling, from communications to material composition tags, isn’t working, the challenge feels more chronic than acute.
In an admittedly very narrow sample conducted by the European Commission in mid-June, it emerged that nearly 40% of product labels did not accurately capture the fibre mix that made up their constituent materials. The issues ranged from incorrectly calculating the share of fibres in blended fabrics (i.e. overstating the amount of cotton versus the amount of polyester in the total mix) to declaring that products were made of more expensive fabrics than they actually were.
To outside observers (and no doubt to government agencies and NGOs as well) this discrepancy between what fashion says, in advertising and on wash care labels, and what it can back up could easily feel like duplicity. After all, what other reason is there to declare something that might induce a shopper to buy a product if that declaration isn’t backed by data?
The reality, in The Interline’s experience, falls much closer to the prosaic: disconnected systems and processes, whereby the intent is for a product (or part of a product) to be made from a preferred material, and marketing is based on that intent, while multi-tier sourcing, and the evolving practicalities of procurement, head off in other directions.
While there are, no doubt, some brands purposefully marketing attributes they know their products don’t really have, in a lot of cases the delta between what a brand communicates and what it actually knows arises from a lack of internal knowledge. That absence is, naturally anathema to the goal of transparency, but it’s often accidental rather than wilful – at least to some extent.
Unsurprisingly, there’s a large cohort of solutions aimed at delivering sustainability and / or traceability that promise to address that knowledge gap. Supply chain mapping is a popular software category for a reason, and the discipline of lifecycle impact assessment is as much about providing a reliable bedrock for communications as it is about compliance.
But these tools are all pointed at the problem of labelling as it exists today, and their value is measured in how much more they can allow you to prove in established domains. And labelling is now barrelling towards a novel fast-moving frontier: the use of AI.
We won’t rehearse the subjects of last week’s newsletter, which examined the machinery of programmatic ad creative, and looked at the consumer backlash that obviously AI-generated images in advertising can create. Instead, we encourage our brand and retail readers to quickly get up to speed on the differing approaches to how the use of AI in advertising and eCommerce will be regulated.
From the 2nd August this year, the EU will, under Article 50 of the AI Act, begin to enforce its Code of Practice on Transparency of AI-Generated Content. As part of this roll-out the bloc has released a set of common icons that it’s encouraging companies to use as part of their labelling of content that is either fully AI-generated or partly AI-edited or enhanced.
The icons fit three scenarios that are intended to feel distinct but, to The Interline’s mind, have a lot of overlap and uncertainty. A general “AI” label can be applied where AI was non-specifically involved in the creation of the content; an “AI generated” label is intended to be placed on content that was created end-to-end with AI; and an “AI modified” label is geared towards content that began life as other media (i.e. a photograph or a piece of pre-existing art) and that was then altered with the use of generative AI tools.
The language used by the EU Commission in separating these labels is similarly murky. In the realm of images and video, it uses the term “deep fakes” extensively, to refer to “AI-generated or manipulated […] content that resembles existing persons, objects, places, entities or events”. It seems, to The Interline, that it would relatively straightforward to mount a defense to this definition, stating that an AI-generated image of a human model does not fall under this aegis, because that image depicts a non-existent person.
To be clear: this labelling scheme would capture the kind of license-stretching, likeness-appropriate case we wrote about recently, where a brand or retailer uses existing (properly licensed and compensated) photography of a real human model as the input to an AI workflow, with the output looking like that model took part in shoots they did not.
What it would not capture, at least by our reading, is the incredibly buoyant market for generative AI tools that place real clothes on fictional but close-to-photoreal-looking people.
Now, The Interline is not giving legal advice here. And therein lies the big concern: if a fashion brand were to publish an advertisement that contained a fully generated model, or a set of models, and a regulator was to complain that the output resembled a real and pre-existing person, the brand would need to have a rigorous chain of custody and accountability for how the final pixels were created.
And this is where current AI workflows and workspaces need to rapidly evolve – either as a result of technology providers baking full history and metadata into every node of their workspaces, or because brands maintain their own governance and their own records.
This feels, for the moment, like a relatively low-pressure step to take in the EU. But regulations in New York are much tighter, with the “synthetic performer” law requiring disclosure in any instance where a human-looking character is depicted, which was generated by AI.
If we picture the typical AI-native workspace, where both early-stage creative and late-stage content workflows are stuffed together and make use of the same models and techniques, it’s quickly become essential for every visual asset that’s likely to have a downstream use to be accompanied by a full track record of the models, prompts, and other inputs (across modalities) that were used to create it.
There is, though, a third option – beyond brand governance and technology vendor compliance tools. And it, too, has its roots in the hedging language and the strange provisions of the EU AI Act.
For AI-generated text (for our purposes advertising copy or product detail page content), AI labelling is considered mandatory along the same lines as it is for images… except where the text has “undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content”.
Or, in other words, fashion and other industries could choose not to architect full accountability into their AI workflows at the tool level, and instead shift the burden to the creatives who sign off on advertising content – making them a kind of “accountability sink”.
This, obviously, would not be a very wholesome outcome. But if the way fashion has approached environmental labelling is any indication, The Interline suspects we’ll see more evasiveness and scape-goating where AI content is concerned, before the machinery of actual compliance is installed.