OpenAI has added virtual try-on to ChatGPT. Users can take or upload a selfie and generate images showing how clothing and accessories look on them before buying. The feature is a small addition to a shopping surface that has been expanding all year, and it changes something structural: when a model can show a shopper what an item looks like on their own body, the product page stops being the primary place where a purchase decision is made. Retailers that have optimized for on-site conversion now have to optimize for being the source a model trusts.
Try-On Becomes a Search Surface
Virtual try-on is not new. Google introduced a version for women's tops in 2023, extended it to dresses the following year and to shoes after that. What is new is the placement: inside a conversational assistant that also answers comparison questions. Users can upload a selfie to see items on themselves before purchase Retail Dive report, which turns a visualization tool into a step in the discovery path rather than a destination.
That shift has a measurable consequence. When try-on lives on a retailer's site, the retailer controls the assortment shown. When it lives in an assistant, the assistant chooses which items to render, and it chooses based on what it can verify about fit, material and availability.
The Selfie Input
A selfie is a high-intent signal. A shopper who uploads one has already decided the category is relevant and is now deciding between items. That is the moment when structured fit data, size guidance and return policy matter more than brand storytelling.
What Changes for Product Pages
Three things stop working. Lifestyle photography loses relative value because the shopper can generate a more relevant image. Vague size language loses value because the model needs numbers. And undifferentiated descriptions lose value because the model cannot quote them as reasons.
Gap has described AI-assisted shopping as the next interface shift after the website Interface shift, and the practical reading is that product pages must become machine-readable specifications first and marketing assets second. Pages built the other way around will be summarized badly, or not at all.
The Evidence a Model Needs
For a try-on answer to be useful, the model needs measurements, material composition, care instructions, size charts tied to body measurements, and a return policy stated in plain terms. It also needs those facts to be consistent across the retailer's own pages, marketplace listings and third-party sources.
Consistency is where most catalogs fail. TikTok has pushed AI-powered discovery alongside one-click checkout Discovery and checkout, and platforms that collapse discovery into a single step depend on clean attributes; a model that encounters three different fabric descriptions for the same SKU will either pick one arbitrarily or skip the item.
Fit Data Is the Moat
Return-reason data is the most valuable fit asset a retailer owns, and almost none of it is published. Aggregating "runs small" and "sleeve too long" into a structured size-guidance block is unglamorous work that directly improves how a model represents the product.
Where Retailers Will Lose Visibility
The first loss is in comparison answers. If a model can only verify fit for three of ten brands, it will present three. The second loss is in accessory attachment, because a model that cannot confirm compatibility will not suggest the add-on. Holiday spending is splitting between value seekers and premium buyers Spending outlook, and both groups ask comparison questions, so visibility loss cuts across price tiers.
The third loss is quieter: brand terms. A shopper who asks for "a wide-leg jean that fits a short inseam" may never see a brand name at all. The answer is a set of items, and the brand only appears if the model has a reason to name it.
Google's Head Start
Google has years of try-on iteration and a shopping graph behind it. OpenAI is catching up on features, but the competitive question for retailers is not which assistant wins; it is whether their catalog is legible to both. Legibility is portable, and that is the argument for fixing data once.
A 90-Day Content Plan
Days one to thirty: audit the top hundred SKUs for measurement completeness and size-chart consistency across channels. Days thirty-one to sixty: publish structured fit guidance built from return reasons, and rewrite the five most-quoted product descriptions so each contains at least three verifiable facts.
Days sixty-one to ninety: measure answer share rather than rankings. Ask the assistants the twenty questions your customers actually ask, record whether your brand appears, and track the change month over month. Gen Z and Gen Alpha research on social and AI surfaces before buying Shopper research, and those cohorts are the first to ask assistants instead of searching, which makes answer share a leading indicator rather than a vanity metric.










