What shipped this week
OpenAI added a try on button to product listings for clothing and accessories inside ChatGPT. Shoppers browsing an item can tap the button, upload or take a selfie, and ChatGPT generates an image of them wearing the product. Users can save favorites or organize finds into folders inside ChatGPT's library, and the feature works on both the website and the mobile app. OpenAI framed the launch as a straightforward addition to how people already research purchases in the chat window.
The company did not publish technical detail on the underlying image model, how selfies are retained, or whether the feature works the same way for every merchant's product photography. That gap matters more than it might look. Try on quality depends heavily on clean, well lit, correctly tagged product images, and OpenAI has not said what happens when a merchant's catalog does not meet that bar, which for most retailers outside a handful of digitally mature apparel brands is the normal case rather than the exception.
A narrower bet than checkout
Try on lands just months after reports, citing OpenAI's own internal data, that Instant Checkout is shifting away from completing purchases inside ChatGPT toward routing users to connected merchant apps instead. The reported reasoning was blunt: plenty of users researched products in ChatGPT, but few actually bought there, and running a marketplace with live inventory, pricing, refunds, fraud, and tax compliance across thousands of merchants turned out to be a heavier operational lift than OpenAI wanted to own directly.
Try on fits a different, smaller claim. It does not ask OpenAI to own a transaction, hold payment data, or guarantee inventory accuracy. It asks OpenAI to generate a convincing image, which plays to the strength of a frontier model lab far more than reconciling SKU level inventory across merchants does. Reading the two moves together, OpenAI looks like it is retreating from the hardest, least differentiated parts of commerce infrastructure while keeping the parts that showcase its actual model capability.
The conversion case, and its limits
OpenAI's push into try on has real data behind it, even if none of it comes from OpenAI's own usage yet. A 2024 Perfect Corp survey found brands offering virtual try on saw sales conversion rates more than double, and 38.4 percent of shoppers surveyed said virtual fitting rooms were the closest substitute to the in person shopping experience they were missing online. That is a credible economic argument for apparel and accessories sellers to care about this feature regardless of which company ships it.
The limit is that this data describes retailer owned try on tools, not a third party chat assistant generating the image on a retailer's behalf. Whether a ChatGPT generated try on image converts at the same rate as a brand's own fitting room tool, built on that brand's own photography and sizing data, is an open question OpenAI's launch post does not answer. Treat the doubled conversion figure as evidence the category works, not as evidence this specific implementation will replicate it.
Catching up to Google, not leading it
Google has run virtual try on since 2023, starting with tops and expanding to dresses in 2024 and to shoes more recently. OpenAI's launch brings ChatGPT roughly to parity with a feature Google has had years to refine across a far larger catalog of participating retailers through its shopping graph. That context should temper how much credit ChatGPT gets for novelty here. This is a capability gap closing, not a new frontier opening.
For retail technology leaders, the practical read is that try on is becoming table stakes across every major AI surface a shopper might use, not a differentiator any single platform can hold for long. Google had it first, OpenAI has it now, and Meta is reportedly testing its own shopping research tools in the same direction. Plan your product imagery and catalog data investments assuming every major AI assistant will eventually want the same clean inputs, rather than betting on one platform's integration as a durable advantage.
What retailers already in the funnel should expect
David's Bridal set up shop on Shopify's Agentic Storefronts for ChatGPT and Microsoft Copilot in April, letting shoppers filter its assortment by silhouette, and Etsy launched its own ChatGPT app the following month. Both retailers are exactly the kind of merchant this try on feature is built for: apparel heavy, photography rich, and already willing to treat ChatGPT as a discovery and research channel even without full checkout integration. Expect more apparel and accessories brands to follow their lead now that try on gives ChatGPT a reason to send shoppers further down the funnel than a product description alone ever could.
If you run commerce technology at a retailer with a presence on ChatGPT, assume this feature raises the bar on what your product imagery pipeline needs to deliver. Inconsistent backgrounds, poor lighting, or missing size and fit metadata will show up as bad try on results attributed to your brand, even though the generation happens on OpenAI's infrastructure. That is a reputational risk sitting outside your own stack, and it is worth a short internal audit before your catalog becomes someone else's try on demo.
The roadmap implication
Taken together, OpenAI's retreat from checkout and its advance into try on describe a company sorting commerce capabilities into what it wants to own versus what it would rather leave to merchants and platforms. That sorting logic is worth copying internally. Any AI feature you are building for shoppers should get the same test: does it require you to own operational risk you are not staffed for, or does it play to a capability your team or model partner genuinely has.
Expect the next round of AI shopping features from every major lab to follow this same pattern, lightweight, model driven experiences like try on and research, paired with a retreat from the operationally heavy parts of transactions. Build your integration roadmap around supplying clean data to those lightweight layers rather than waiting for any single AI platform to become a full checkout partner, because the evidence this quarter points the other way.
