OpenAI has added a virtual try-on feature to ChatGPT, letting users upload a selfie or full-body photo and see how an item of clothing would look on them. The company said on 1 October that a "try on" button now appears wherever a clothing product listing shows up in a chat, worldwide, and that the images come from its ChatGPT Images 2.5 model, released last month.
The feature is not limited to products ChatGPT surfaces. Users can upload a screenshot of a garment from anywhere and ask to see it on themselves. OpenAI also launched a library where people can save items they like, stored alongside the try-on images they generated.
That combination, a product feed, a visual try-on and a saved list, is the shape of a shopping app rather than a chat assistant, and it places OpenAI in the part of retail where decisions are made.
OpenAI Has Been Working Towards The Checkout
The try-on button follows a longer push into commerce. OpenAI launched Instant Checkout in September 2025, built on an Agentic Commerce Protocol developed with Stripe, which let users buy from Etsy sellers and Shopify merchants without leaving the chat, with merchants paying a fee on completed purchases.
That effort did not work out as hoped. Instant Checkout ended up not performing well, according to TechCrunch's account, and OpenAI's emphasis has shifted towards discovery, where it also sells advertising to merchants inside chats.
Discovery may be the more valuable position anyway. The company that shapes which three jackets a shopper considers has more influence over the sale than the one that processes the payment, and it carries none of the fulfilment or returns cost.
Where Retailers Sit In This
For brands, an AI assistant that recommends products and shows them on the customer is a new intermediary between them and their buyers. It resembles the arrival of search advertising or marketplace listings, where visibility came to depend on someone else's ranking.
Retailers will want to know how products get selected for these results, whether paid placement affects what appears, and what data they receive about the shoppers who tried their items on. None of that is public yet.
Does Virtual Try-On Actually Sell Clothes
The commercial case rests on two problems in online apparel: low conversion and high returns. Online clothing returns run at roughly 30% to 40%, against ecommerce conversion rates of under 2%, and much of that comes down to customers being unable to judge fit and look.
A study published in July by DRESSX, which sells virtual try-on technology and therefore has an interest in the result, reported that shoppers who used try-on added items to cart at 11% against 4% for those who did not, across 1.2 million shoppers in 216 countries. View-to-purchase conversion was around 50% higher.
Those numbers describe correlation rather than cause, since a shopper who bothers to try an item on is already more interested than one who scrolls past. The retention figures in the same study, 44% of try-on users still active after 30 days against 1% of others, are large enough to suggest the comparison is between quite different groups of people.
Fit Is Still The Hard Part
A generated image shows how a garment looks, not how it fits. Current systems render a plausible picture rather than modelling a specific body against a specific cut, which is where most returns come from, and nothing in OpenAI's announcement suggests it has solved sizing.
Retailers hoping this will cut returns should test it on their own data before counting the saving.
The Privacy Problem Lands On Everyone
The feature asks people to upload photographs of themselves, and by default those images are used to train OpenAI's models unless the user opts out. That alone deserves attention from anyone considering similar features.
Retailers have been here before. Virtual try-on tools for makeup, eyewear and jewellery drew a wave of class actions in Illinois under its Biometric Information Privacy Act, which requires written notice, consent and a published retention policy before collecting biometric data, with damages of $1,000 per negligent violation and $5,000 for reckless ones. Estée Lauder, Louis Vuitton, Pandora and others faced suits, with mixed outcomes depending on whether courts treated the scans as biometric identifiers.
A brand whose products appear in a third party's try-on tool should establish who holds the images, under what consent, and where liability sits if a customer later objects. The same questions follow any system where AI agents act on a person's behalf using their data.
Google Got There First
OpenAI is not leading this. Google added virtual try-on to Search in 2025, using the same upload-a-photo approach, and has been building shopping features into its AI results for longer. Amazon has run try-on tools in its app for years.
The difference OpenAI is betting on is habit. If people already describe what they want to ChatGPT in conversation, adding the ability to see it on themselves keeps them there rather than sending them to a search engine or a retailer's site. Whether fashion discovery, long held by Pinterest, Instagram and Google, moves to a chat window is the open commercial question.
What Brands Should Do Now
The practical steps are modest but worth taking early. Brands should check how their products appear in ChatGPT's shopping results and whether the imagery and descriptions being used are their own. They should decide whether they want their catalogue available to these systems at all, and on what terms.
They should also treat any try-on feature they build themselves as a privacy matter first and a conversion tool second, with explicit consent, a retention policy and a clear answer on whether uploaded images train any model.
The Assistant Is Becoming A Shop Window
The try-on button is a small feature, and the generated images will not convince anyone they have seen a real fitting. Its significance is in where it sits: inside the conversation where someone decides what to buy, owned by a company that also sells the advertising alongside it.
Retailers spent two decades learning to compete for position in search results. The question now is whether a comparable contest is forming inside AI assistants, and on what terms brands will be able to appear there.
