Fashion week is debating AI in the studio. Shoppers already brought it to the store.

Vogue Business spent fashion month backstage asking designers what they actually think about AI, and came back with three answers instead of one. Some brands are using it as a genuine creative partner. Some use it heavily, but only behind the scenes, on mood boards, mock-ups, and contracts. Some refuse it and stake the brand on work no machine touched.

The disagreement is real and the detail is worth reading. Marie Lueder made her show invites with AI and found it "wasn't faster or easier." Demna, who teased his Gucci debut with AI-generated images, compared the holdouts to the retailers who refused e-commerce in 2008. Paul Billot designed an entire debut collection by feeding a poem into an AI model, and refuses to call the technology a tool or a partner: "I like to say that it is a material, because it has some constraints and properties to embrace." Dimitra Petsa of Di Petsa argued the other way, that AI will never feel when a design is relevant, and predicted shoppers will swing back toward things they can physically touch.

Every one of those is a position on how clothes get made. Not one of them is a position on how clothes now get found. That second question is moving faster, and nobody backstage was asked about it.

The shopper already picked a side

The Ryder 2026 E-commerce Consumer Study puts AI-assisted shopping tool use at 64% of shoppers. Search and shopping assistants lead at 45%, review summaries at 37%, image search at 33%, and virtual try-on at 31%.

The more pointed number is about trust. In the consumer survey behind The State of Fashion 2026, from McKinsey and The Business of Fashion, more than 40% of consumers said AI answers felt more reliable than paid ads. Only 15% said less reliable.

That is not adoption, it is a preference. Shoppers would rather hear about your collection from something you did not write than from something you paid for. So the first description of your work a buyer reads is increasingly one you never approved, assembled out of whatever a machine could find about you.

What the runway sells does not survive the summary

The sharpest evidence for this came from a designer who was not looking for a marketing lesson. Before he built a collection with AI, Paul Billot spent his MA thesis documenting what AI cannot do: understand fabrics and the weight of garments. He then turned that gap into a design method. Fair enough in a studio. On a product page it is a commercial problem, because weight and drape are not decoration in fashion. They are the product.

A shopper standing in a store settles this in two seconds by touching the sleeve. A shopper asking an assistant settles it with whatever words exist about that sleeve somewhere a machine could reach. If those words do not exist, the assistant does what people do when they lack information. It moves on to a product it can describe.

The runway sells the feeling. The recommendation sells the facts. Only one of them travels.

What actually travels

For a fashion brand, the things that survive being summarized by an assistant are narrower than most teams assume, and more specific.

• Fabric composition with weight. "100% cotton" is a fact. "310gsm brushed loopback cotton" answers the question the shopper was actually asking, which is how heavy this thing will feel.
• Measurements per size, as text. Chest, length, and inseam for every size. A size chart baked into an image is invisible to the thing reading your page, and it is a common gap on apparel sites.
• Model height and the size worn. Two short facts that turn every photo on the page into a scale reference.
• Fit language from reviews. When forty people write that it runs small in the shoulder, that phrase is more quotable than anything in your product copy, because it did not come from you.
• Origin and who made it. Named mill, named workshop, run size. Checkable, and increasingly what the buyer is weighing.
• The return terms in plain words. Ryder found 52% of consumers unwilling to buy without free returns, and 79% who had hit a returns problem that stopped them shopping with a brand again.

What does not travel is most of what a fashion brand spends its money on. The show does not travel. The campaign film does not travel. The styling, the set, the casting, the sequence of looks that made the collection make sense: none of it reaches a shopper whose first encounter with you is four sentences in a chat window.

The refusers have the better story, if it is readable

The most interesting part of this is on the refusers' side. Petsa predicted that doubt about what is real will push shoppers toward what they can touch. The Ryder study found something consistent with that in apparel specifically: over the past year, price fell 12% in its weight on purchase decisions, while materials, environmental impact, and ethical practice gained ground. Two different readings pointing the same direction. The argument buyers are having with themselves is shifting from what it costs toward what it is and who made it.

That is good news for a brand whose entire case is human hands. It only converts if the case is legible.

"Crafted by hand" delivered in a show note and a mood film is a claim no assistant can repeat, because there is nothing in it to check. The same claim written on the product page as the workshop's name, the technique, the hours, and the size of the run becomes exactly the kind of detail a recommendation will carry, because it can be defended. The refusal to use AI in the studio and the need to be readable by AI in the market are separate decisions, and brands keep collapsing them into one. The designers with the most checkable facts on their side are often the ones burying those facts in a paragraph of atmosphere.

Before the next round of shows

Miuccia Prada, backstage in Milan, said AI is already here and that it falls to the industry, governments, and other organizations to control it. On regulation that is a reasonable position. On the commercial side there is no waiting period. The next round of shows starts this month, and those collections will be described to shoppers by systems reading product pages written for a different era of discovery. Whatever the systems can find is the collection, as far as a growing share of buyers is concerned.

The work splits cleanly in two. Make the facts readable, so the machine doing the recommending is quoting you rather than guessing. Then check whether those facts land when a real buyer reads them, because a page can be perfectly machine-readable and still fail the person it was written for.

eLLMo does both halves. We build the structured data and semantic alignment layer underneath a catalog, so the engines doing the recommending can read your fabrics, fit notes, reviews, and availability, and quote them without inventing anything. Then we run test buyers matched to your real customers against the page and return a ranked list of what stops people from buying, in their own words. On apparel that list is almost always a question about fit, feel, or returns that the page left the buyer to guess at. See a live run before the season decides it for you.

The three camps backstage are arguing about the right thing for their studios. The buyer side was settled while they were arguing, and it did not require anyone to have an opinion about AI at all.

*Related Links: AI Has Officially Entered the Fashion Week Conversation (Vogue Business), The State of Fashion 2026 (McKinsey and The Business of Fashion), Ryder 2026 E-commerce Consumer Study, Discovery has left your website, Product pages AI cannot read.*