The dress turns before it sells. A hem kicks as the model pivots, the knit skims the hip and releases, and the chat window shows the one thing a flat product shot has never managed: how the garment behaves on a moving body. This is ASOS Stylist, the app ASOS launched inside ChatGPT on 20 May, and it is the first time the conversational aisle has played video rather than described it. The next contest in AI commerce will not be the answer the model gives. It will be the picture it can show.

The trade press filed it as another AI shopping app, which is where the framing goes slack. Conversational commerce, a chatbot answering “what should I wear to a garden wedding” with text and a grid of stills, is a year old and already crowded. What landed on 20 May is the answer as moving image. ASOS built it with Bambuser, the livestream-commerce firm, whose “Intelligence Layer” turns a video library into data a language model can read, retrieve, and play back in real time.

Why this matters is a question about cloth, not software. A still photograph freezes drape at one flattering instant and hides everything the fabric does next. ASOS’s head of digital product, Melissa Lim, says the company’s A/B tests found that video gives “more nuanced information about fit” — the swing of a hem, the way a knit pulls or sits clean across the shoulder. Motion is where a garment confesses its weight.

A garment has always sold best in motion, and the photograph was the interruption.

Charles Frederick Worth had live models walk his designs through his Paris salon in the 1860s, because a gown had to move before a client could judge how it fell. Everything since flattened that instinct into the catalogue page and then the e-commerce hero shot, a single frame chosen to sell the fit, not report it. ASOS Stylist is Worth’s premise rebuilt for the answer engine: let the customer watch the thing walk before she commits.

That relocates the prize. The barrier is not the chatbot, since OpenAI opened that storefront to any brand that can build an app inside it. The barrier is the video catalogue rendered as structured, machine-readable data, the unglamorous metadata work we have flagged before as the real constraint on AI fashion imagery. A brand whose clips sit on a server as marketing no model can parse will show up in the chat as text and a flat grid. Film the catalogue so the machine can read it, and the brand shows up moving.

The chat window stopped being a transcript and became a screen. Every AI-shopping pitch so far has tuned the sentence the model returns; the sentence is now the cheap part. ASOS, with 17 million active customers it wants to grow, is betting that the shopper typing “something for a beach wedding in July” would rather watch the dress turn than read about it. The brands that filmed their catalogues so a machine can read them will own that turn. Everyone else will be narrating clothes to people who can already watch them move somewhere else.