Mandatory AI labeling, ever more demanding marketplaces: in fashion e-commerce content, technology is only valuable if it produces better, not just more.
Since August 2026, a line has moved quietly in the production of fashion content: the obligation, posed by the European AI Act, to clearly indicate that a visual has been generated or substantially modified by artificial intelligence. There has been a lot of commentary on the principle. We have hardly looked at what changes, in concrete terms, for the teams who produce the thousands of images and videos every day that make up an e-commerce business.
But it changes a lot. Because it transforms an aesthetic question into an operational question: from the moment “generated by AI” becomes a mandatory mention, each brand must know precisely what its technology does on each asset — and assume responsibility for its readability. This is no longer a principled debate about AI. It’s a method subject.
Transparency moves the cursor from “how much” to “how”
For two years, the conversation revolved around volume: producing faster, more formats, for more markets, at lower cost. Generative AI promised exactly that. The labeling obligation reshuffles the cards, not because it prohibits anything, but because it makes visible a distinction that many had left unclear: that between an image created by an algorithm and a real image, simply optimized by technology.
These two approaches do not carry the same regulatory risk, nor the same compliance burden. A fully generated visual must be reported, traced, documented. An authentic photograph whose AI has accelerated the clipping, the variation of formats or the colorimetric harmonization involves another logic. For an e-commerce department, the difference is not philosophical: it is accounting, legal and organizational.
What labeling imposes, in practice
Compliance is not decided at the end of the chain. It is built in the workflow. Concretely, three areas are essential for brands that want to keep up the pace without exposing themselves.
The first is traceability. We must now be able to say, for each asset, what intervention the AI had – and to prove it. This requires content governance: naming, metadata, production history. Organizations that previously managed their visuals as scattered files are finding they need to manage them like a documented asset.
The second is multi-channel consistency. Amazon, Zalando, Farfetch, brand sites: each has its formats, its specifications, and soon its own requirements for displaying mentions. Producing at scale while remaining compliant on each channel becomes a problem of industrialization, not of creation.
The third, the most strategic, is the preservation of identity. In luxury and premium, the image is not a support: it is a part of the product. The challenge is therefore not to choose between AI and humans, but to place each where they are best. AI excels in variation, volume and standardization. The human element remains irreplaceable in the artistic direction, the judgment, the fidelity to the universe of a house. It is this division that the new rule basically makes more readable.
Lavorare meglio, non di più
This is where a conviction, for me, is confirmed rather than discovered: artificial intelligence is only of interest, in our profession, if it makes us lavorare meglio, non di più — work better, not just more. The opposite temptation exists. It consists of seeing AI as a machine for stacking visuals, faster and cheaper, to the point of saturating the channels with interchangeable images. Fast fashion has already taken this path. The result is known: quantity without memory, which does not build any brand preference.
The context of the sector makes this discipline all the more decisive. The personal luxury market has lost several million consumers in two years, and those who remain are more demanding and more volatile. Every conversion point counts, every image carries part of the promise. In this context, automating repetitive production only makes sense if the time and budget thus freed up are reinvested where the difference comes into play: the campaigns, the editorials, the content that makes a brand exist rather than simply exposing it.
Labeling is not a constraint, it is a revealer
From a distance, the transparency requirement looks like an additional layer of complexity. Seen up close, it acts as a revealer. It separates organizations that know what their technology does from those that have adopted it without method. It values brands that are able to say — because they built it that way — that their images remain real, produced by professionals, with technology only intervening to serve quality and scale.
The ongoing transformation is therefore not a matter of tools. It is a matter of governance and choice. The right question for a brand manager is not “should we use AI?” — the market has decided — but “is my content production organized so that AI makes me work better, and not just produce more?”. Those who know how to respond will make content a lever for growth. The others will discover, one legal notice after another, the cost of the volume without method.
