ChatGPT Just Made Virtual Try-On Table Stakes. It Also Proved Why Merchants Need Their Own.

When the feature appears on an eligible clothing or accessory listing, shoppers can upload a selfie and generate an image of how the product could look on them. ChatGPT also lets shoppers save products and, for some merchants, complete checkout without leaving the app. OpenAI’s release notes make the direction obvious: ChatGPT does not just want to answer shopping questions. It wants to own more of the shopping journey.

That is huge news for virtual try-on.

It is even bigger news for merchants.

The core argument in my upcoming book, Agentic Commerce’s Dirty Little Secret, is that answer agents are becoming a new front door to commerce. They will help shoppers discover, compare, visualize, save, and increasingly purchase products. Merchants need to participate in that ecosystem. But they should not confuse participation with a winning strategy.

Being agent-ready is table stakes. Protecting the castle is the strategy.

The agent may be the road to the merchant. The merchant’s own site still has to be the place where the relationship, the data, the brand, the basket, and the next purchase are earned.

Virtual try-on has crossed the line from novelty to expectation

The market was already headed here. Zara, Walmart, Amazon, Google Shopping, Diesel, Jil Sander, and Kmart have all made meaningful public moves into virtual try-on. One launch can be a pilot. Five or six launches across mass retail, premium fashion, marketplaces, and major discovery platforms are a pattern.

Now ChatGPT has joined the list.

The significance is not that every generated image will be perfect. OpenAI is appropriately clear that try-on images may not represent the actual product, appearance, fit, or size precisely. Shoppers still need product details, measurements, and return policies. OpenAI’s shopping guidance says exactly that.

The significance is that the behavior is being normalized.

A shopper can ask ChatGPT for a dress, see products, upload a photo, and begin answering the question that matters most in apparel: “Would that look like me?” That decision no longer has to happen on a merchant’s product page.

And that is the danger.

Three answers from one merchant

I tested ChatGPT with a completely normal shopping query:

“Show me some fall midi dresses that would be good for a weekend BBQ with friends.”

Three results appeared.

All three were from Nordstrom.

Why?

ChatGPT has access, in theory, to an enormous universe of products and merchants. Yet it compressed that universe into three choices from one retailer. Nordstrom may have had highly relevant inventory. It may have stronger structured product data, greater authority, better availability signals, or simply been the best fit for that specific query. Or is it reflecting paid placement? The point is not that Nordstrom did anything wrong.

The point is the math.

Imagine you are a $20 million GMV merchant with 100 excellent midi dresses. Several were specifically designed for BBQs at casual fall gatherings the shopper described. What are the odds that your product makes it into a three-result answer when the agent has already decided that one large retailer can satisfy the request?

This is the selection problem at the heart of agentic commerce.

A structured feed makes a merchant eligible. It does not make the merchant visible.

AI agents do not recreate a merchant’s aisle. They compress a huge market into a tiny answer: three products, a short comparison, a few purchase links. Every merchant is fighting to become one of those few choices. The more capable the agent becomes, the more important that compression becomes.

That is why agent-readiness alone is not a growth strategy. It is the price of admission to an extraordinarily competitive top of funnel.

The general-purpose experience is still not the best shopping experience

I tested ChatGPT’s new try-on experience. It took roughly 60 seconds to produce a single-item composition.

That is not a knock on a brand-new launch. This is hard technology, and it will improve quickly. But it is a useful reminder that a general-purpose answer agent is not automatically a great merchant experience.

In my testing, studioGPT produces a composition in roughly 6–10 seconds. More important than the speed, though, is what the experience is designed to do. It is not limited to placing one isolated item on a shopper. It can help visualize a selected outfit and keep that visual decision connected to the merchant’s real catalog, product variants, availability, complementary items, and purchase path.

See the intended merchant-owned flow in the studioGPT virtual try-on demo.

That distinction matters because shoppers do not buy apparel one disconnected product at a time. They want to know whether the top works with the skirt, whether the shoes complete the look, and whether the outfit makes sense for the occasion. The commercial opportunity is not merely a picture of one dress. It is a complete look, visualized together, with the ability to add the pieces to one cart.

And shoppers will increasingly expect more control than one generated image. They will want multiple poses, angles, and contexts for the same selected outfit: front view, side view, seated, walking, or an image that better fits the occasion for which they are dressing. That is when virtual try-on becomes a real decision tool rather than a clever composition.

There was another interesting wrinkle in my test. I searched ChatGPT for maxi dresses and did not see the promised Try-On icon in the results. Again, it is a new rollout, so this is not surprising. But it is still telling. Product launches, especially those from a trillion dollar company, do not become customer experiences just because the feature exists in a release note.

For merchants, the lesson is straightforward: do not add virtual try-on because it is fashionable. Add it where it is fast, visible, useful, and connected to purchases that keep clients onsite.

Meta and Shopify make the same point from another direction

A LinkedIn post by Dan Sheard highlighted another development that belongs in this conversation: Shopify now supports direct checkout on eligible Meta surfaces. A shopper can discover a product there, complete checkout without entering the merchant’s website, and have the order arrive in Shopify as part of normal operations.

The merchant remains merchant of record. The merchant still fulfills the order, handles returns, and supports the customer.

Convenient? Yes.

Neutral for the merchant? No.

Shopify’s own documentation says orders made through Meta direct checkout are attributed to Meta. It also says that third-party analytics pixels—including standard and custom client-side pixels—do not fire when checkout happens on Meta’s surface. That makes the merchant a fulfillment center. Shopify’s Meta direct-checkout documentation also notes that eligible merchants can opt out and redirect shoppers to their own stores instead.

That is the agentic-commerce pyramid in action.

The merchant gets distribution and a sale. The platforms above it get the discovery surface, the attention, the behavioral signal, and a privileged place between the shopper and the brand. Shopify powers the commerce layer. Meta owns the feed. Google owns the search. ChatGPT wants to own the answer and more of the action after it.

None of this is a conspiracy. It is the business model. And a GOOD one!

Be agent-ready. Then make your own site irreplaceable.

Merchants should make their products legible and transactable everywhere that AI shoppers begin:

• Clean product data
• Current inventory and pricing
• Accurate variants and attributes
• Agent-readable policies
• Structured feeds
• Checkout paths that work across channels

Do all of that.

But do not stop there.

A merchant’s own site must be the better place to continue the decision. It should be the best place to search the full assortment, discover visually, ask product-specific questions, create an outfit, see multiple try-on options, understand the product facts, add complementary items, and become a known customer.

That is the Protected Castle.

The merchant that treats agentic commerce as a magic distribution portal will lose traffic, branding, customer data, and margin one transaction at a time.

The merchant that treats outside agents as roads—and builds a truly superior onsite discovery and try-on experience—has a chance to turn that traffic into a compounding customer relationship.

ChatGPT’s VTO launch validates virtual try-on. It also validates the urgency of owning the experience around it.

The agent can introduce the shopper.

Do not let it take complete ownership of your client.