ChatGPT Lets Shoppers Try Clothes On Without Leaving the Product Card

Browsing for clothes online has always had a gap between “this looks nice” and “this would look nice on me.” Shoppers fill that gap with guesswork: they study the model’s photos, read the size chart, scroll through reviews and hope for the best. ChatGPT now tries to close that gap inside the same window where shoppers find the product in the first place.

The change is small on the surface, a single button on a product card. But it moves part of the evaluation step into the place where discovery already happens, and that has implications for anyone who sells fashion or accessories online.

What OpenAI added

According to OpenAI’s release notes from October 1, 2026, clothing and accessory listings in ChatGPT now carry a “Try on” button. Tapping it asks for a selfie, which can be taken on the spot or uploaded from the device. ChatGPT Images then generates a picture of the shopper wearing the item.

The selfie is kept as a reference photo, so later try-ons don’t require a new upload. Users can replace or delete that photo under Settings, then Personalization, then Reference photos.

The same release adds a second feature that is easy to overlook: shoppers can save products to Favorites or organize them into folders in the ChatGPT Library. Both the try-on button and the saving tools are listed as available on mobile and web.

What the announcement does not include matters just as much. It gives no exact publication time, no conversion figures and no data on how many shoppers use the feature or how it affects purchases.

How the flow works in practice

Put together, the experience described in the release notes looks like this:

A shopper finds a clothing or accessory item in a ChatGPT product listing.
They tap “Try on” on that listing.
They take a selfie or upload one.
ChatGPT Images generates an image of how the item could look on them.
The reference photo is stored for future try-ons, and the shopper can manage or delete it in Settings.
If they aren’t ready to decide, they can save the product to Favorites or put it in a folder in the Library.

Nothing in the notes says the shopper has to leave ChatGPT during any of these steps. That is the real change: discovery, visual evaluation and saving for later all happen in one interface.

You control the reference photo

Because the feature relies on a photo of the user’s own face and body, the control over that photo is worth noting. The release notes say the reference photo can be changed or deleted through Settings. That puts the stored image under the user’s management rather than leaving it as a permanent part of the shopping flow.

For anyone writing about or recommending the feature, the editorial takeaway is simple: shoppers should know where the photo lives and how to remove it. The release notes cover those two points, but they don’t go into how the image is used beyond try-ons, so we won’t speculate about that here.

A picture, not a fitting room

The most important limit is in what the feature actually shows. A generated image suggests how a piece might look on someone. It is not a measurement. It doesn’t tell a shopper whether a jacket will be tight across the shoulders, how a fabric will drape, or whether a size runs small, and nothing in the announcement claims it does.

The release describes how an item could look, not a physical fitting guarantee. That distinction should shape how both shoppers and merchants treat the result:

For shoppers: the image is a visual impression, useful for judging color, style or overall look. Sizing charts, material details and reviews remain the evidence for fit.
For merchants: a generated try-on doesn’t replace accurate specifications. If anything, clear sizing information becomes more important, because it is the part the image can’t supply.

This section is our editorial interpretation, not something OpenAI has stated. Until now, the path from finding an item to imagining it on yourself usually involved several steps and several tools: a search, a store page, a gallery of photos, perhaps a separate app or a video. Each transition is a chance for a shopper to drop out of the process.

With a try-on button on the product card, those steps collapse into one. The listing becomes more than a description of a product. It becomes the starting point for a second stage of evaluation, where the shopper is no longer asking “what is this?” but “how would this look on me?”

That doesn’t prove the shopper will buy more. It does change the questions merchants should be asking about how their products are presented, even before there is evidence of any commercial effect.

What this means for merchants and SEO teams

The practical recommendation here is modest and doesn’t depend on how the feature performs: review whether your product images clearly communicate what you are selling. A quick audit can cover a few points:

Do the photos, descriptions and variant details all refer to the same product? A mismatch between a listing’s photo and its stated color or style makes any evaluation harder.
Are color names unambiguous? Vague or inconsistent color labels leave the shopper unsure what they are looking at, whether or not they use try-on.
Is the photography current? Outdated images of a product that has since changed make it harder for a customer to judge the item they would actually receive.
Is the product clearly visible? Images that show the item plainly, on its own and in use, give shoppers more to work with.

These are editorial quality recommendations. OpenAI’s release notes do not establish any new image ranking factors, any rules for which merchants are admitted, or any guaranteed relationship between image quality and visibility. They also don’t explain how every product qualifies for the “Try on” button, so it would be wrong to assume that cleaner images unlock the feature.

Being in the results is not the same as being chosen

A merchant needs to distinguish between whether an item appears in a shopping result and what happens after it appears. Virtual try-on adds another possible interaction, but its availability does not prove that a product will be selected, clicked or purchased.

Those are separate stages, and each can succeed or fail independently. A product can show up and never be tapped. It can be tried on and still be rejected. It can be saved and never revisited. Treating any single stage as proof of the next one leads to bad conclusions.

Keep the metrics separate

Where measurement is available, it helps to track the following as separate things:

Product exposure: the item appearing in a shopping result
Try-on interactions: a shopper using the feature on the item
Merchant visits: a shopper arriving at the merchant’s own page
Completed purchases: an actual sale

Avoid counting a generated image as a conversion. It isn’t one. A shopper may use the visualization to reject an item, to compare it with alternatives, or to return to it later. Those are three different outcomes with three different commercial meanings, and a metric that lumps them together will hide what is actually happening.

It also helps to be careful about cause and effect. If sales change around the same time the feature rolls out, that doesn’t establish that virtual try-on caused the change. Seasonality, promotions, stock levels and many other factors move at the same time.

Saved products may stretch the buying journey

Favorites and folders give shoppers a way to come back to products instead of deciding right away. Our inference is that a saved item can belong to a longer research sequence: a shopper tries something on, saves it, looks at other options and returns days later.

The announcement doesn’t show how often people return to saved items, and it doesn’t say whether saved products receive any preferential placement. Those are open questions, and any claim about them would be a guess.

What it does suggest is that teams assessing performance should think about the timing of a decision as well as the first interaction. A same-session click is only one possible signal. A shopper who saves an item today and buys it later would be invisible to a measurement window that only looks at the first visit.

Questions the release leaves open

The release notes answer what the feature does, but several things that matter to merchants and publishers remain unanswered:

How many shoppers will actually use the “Try on” button, and how often
Which products qualify for the button and which don’t
Whether virtual try-on changes click-through, purchase or return rates
Whether saved products are treated differently in later results
Whether image quality has any bearing on visibility

Until OpenAI or independent measurement answers these, they are open questions rather than established facts.

What to do now

For a store or an SEO team, the sensible response is measured. Improve what is in your control and observe what actually happens:

Audit product photos, descriptions and variants for consistency.
Make sizing and material information as clear as possible, since the generated image can’t provide it.
Where you have visibility into shopping interactions, record exposure, try-ons, visits and purchases separately.
Look at decision timing over days rather than only within a single session.
Hold off on strategy changes that assume better rankings, more accurate fit or higher conversion until there is evidence for them.

A new experience, with commercial effects still to establish

The official update expands what shoppers can do inside ChatGPT: discover an item, visualize it on themselves and save it for later. That is a meaningful change in how product discovery works, and it is worth taking seriously.

But the evidence for any commercial effect isn’t there yet. Claims of better rankings, more accurate fit or higher conversion would require evidence beyond these release notes. For now, the useful move is to make product information clearer and watch how shoppers actually behave.