In short. ChatGPT uses a clothing listing and a reference photo to generate a visual try-on preview, while product and fit details remain in the merchant listing.
Definition: Virtual try-on in ChatGPT is an image-generation flow attached to some clothing and accessory listings. Selecting Try on starts from the listing and asks for a selfie or uploaded photo, then returns a generated image showing how the item could look on the person.
Key facts
- OpenAI's shopping help documents a Try on button on clothing and accessory listings, a selfie or uploaded photo as input, a generated try-on image as output, saved reference photos for later try-ons, and availability on mobile and web. It also says product surfacing considers structured metadata such as price and product description.
- OpenAI's commerce onboarding guide says a structured product feed supplies catalog data including titles, descriptions, images, price, and availability. Feed onboarding is currently available to approved partners.
- OpenAI's stable product feed reference requires a direct
image_urlfor the main image showing the selected variant, and says that each variant row carries its own price, availability, URL, and images.
What happens when a shopper selects Try on?
OpenAI's shopping help describes a three-step interaction: select Try on on a clothes or accessories listing, take or upload a selfie, and view the generated try-on image. The same page says that a shopper can upload a picture of clothing or accessories in a conversation and ask ChatGPT to show how the item could look on them. The image is produced with ChatGPT Images, so the result is a generated visual preview rather than a product photograph supplied by the merchant.
Reference photos are retained for future try-ons. OpenAI places the controls at:
Settings → Personalization → Reference photos
The help page says a shopper can change the photo or remove it from that area. It lists the feature as available in ChatGPT on mobile and web.
Which listing data is relevant to the generated image?
A catalog image is the merchant supplied image associated with a product or variant. OpenAI's stable feed reference defines image_url as the required main product image, showing the variant, and asks for a direct JPEG or PNG URL. It also documents additional_image_link as an optional field for comma-separated URLs mapped to additional images. Product and image URLs must be publicly accessible.
The same reference models clothing and other product variants as separate feed rows. Each row can carry its own price, availability, product URL, and images, while a shared group identifier connects selections within one product listing. A merchant can therefore give ChatGPT variant-specific image and stock data through the documented feed structure.
OpenAI's public shopping and commerce documentation connects the listing to the Try on control and documents the catalog image fields. It does not specify which particular listing image or additional media input is passed to ChatGPT Images for a given generation.
What should a merchant check in a product feed?
OpenAI's commerce guide says that product feeds give ChatGPT catalog data needed to index products, understand core attributes, and present product information in shopping experiences. It lists titles, descriptions, images, price, and availability as feed data, and documents file upload and API delivery. The guide recommends sending a complete file-upload feed once a day, followed by updates through the API.
The stable product reference provides the concrete image field and the surrounding listing fields. A minimal record uses a stable item identifier, a product detail URL, a seller, an image URL, price, and availability. The image portion follows this documented pattern:
"image_url": "https://example.com/images/trail-black.jpg",
"availability": "in_stock",
"price": "79.99 USD"
The same reference says legacy virtual_model_link metadata may be accepted from existing feeds but is not used for product discovery. The public documentation therefore supports maintaining accurate product and variant images, while it leaves the internal handoff from a listing to the generated try-on image unspecified.
What can the generated image establish?
OpenAI's shopping help warns that try-on images may not represent the product or the person's appearance exactly. It directs shoppers to the merchant's measurements, product details, and return policy before buying. A generated image can show an appearance preview, while the listing remains the place for the item information used to evaluate the purchase.
Related terms
FAQ
Can a shopper use a clothing photo instead of a product listing?
Yes. OpenAI says a shopper can upload a picture of clothing or accessories in a conversation and ask ChatGPT to show how the item could look on them. The documented listing flow uses the Try on control and a selfie or uploaded photo.
Does the generated image confirm fit or size?
No. OpenAI says try-on images may not represent the product or appearance exactly and do not establish fit or size. The shopper should consult the merchant's measurements, product details, and return policy.
Where can a shopper manage the reference photo?
OpenAI documents Settings → Personalization → Reference photos for changing or removing a saved reference photo. The photo is saved for reuse in future try-ons.
Sources
- OpenAI Help Center, "Shopping with ChatGPT Search", undated
- OpenAI Developers, "Get Started", undated
- OpenAI Developers, "Products", undated
Check how a page reads to answer engines. Open the GEO score analyzer