ChatGPT Product Feed Images: Match Each Variant to the Right Photo

Quick answer

Make a row for each purchasable variant, preserve a unique stable item ID and pair it with the right source photograph. Check the final public image address against the variant and landing page. Have the authorized feed owner validate the complete current schema and eligibility before submitting outside Ciyo.

What changed, and what did not

OpenAI announced Businesses in ChatGPT in beta on November 10, including Business Hub and eligible business capabilities. That announcement is the news hook; it is not evidence that your store is eligible, that demand for this exact query is growing, or that a submitted product will appear to a shopper.

The supported file-upload reference distinguishes Stable from a Draft intended for planning and feedback. Use the current supported schema for the actual integration. This article isolates image matching; it is not a complete feed template, an Ads setup guide or a checkout integration.

1. Build a variant-to-photo worksheet

Start with your approved catalog and authorized originals. Give each row the item ID, selected variant, source-photo filename, intended public image URL and matching product-page destination. Keep the commercial facts under the catalog owner’s control.

The stable reference calls for a unique item ID per item or variant and a main image showing that variant, using a direct image URL. Its upload guidance requires public product and image URLs. Those checks are a starting point, not the entire schema.

2. Review identity before visual consistency

A matching background should not make two different products look like the same variant. Compare each proposed image with the authorized source, checking color, shape, label, included parts and other distinguishing details. If a creative edit changes those facts, return it for correction instead of treating visual polish as approval.

In our deliberately flawed River mugs draft, both variants use the item ID RIVER-MUG and both point to ivory.jpg. The blue mug therefore has a wrong-photo assignment as well as a duplicate identifier. These filenames are illustrative; no public product image or live merchant feed is supplied.

3. Use Ciyo for the review brief

Upload the worksheet or visual contact sheet and state which catalog facts are approved. Ask the Design agent to identify duplicate IDs, wrong variant/photo pairings and unresolved handoff checks. When using real product references, inspect the sources yourself as well; an agent response is not catalog authentication.

The deliverable at this stage is a correction list for the person preparing the feed. Keep it separate from the production data file. A canvas image of a table cannot establish correct schema serialization, public URL access, merchant approval or successful ingestion.

Live Ciyo review: duplicate IDs and a wrong-color assignment

On November 14 at dev.ciyo.ai, Ciyo identified that both fictional rows used RIVER-MUG and that the blue row pointed to ivory.jpg. It proposed distinct example identifiers, the blue photo for the blue variant, and feed-owner approval of the final mapping.

It also flagged that the public image URLs were placeholders. No real URL was tested and no feed was prepared or submitted. We used the identity and mapping corrections; the production feed owner must check the current schema and actual image access rather than treat the agent’s generic domain comment as a verified platform rule.

Copyable variant-image review

Replace the bracketed values with your own approved catalog information. Do not put access tokens or private customer data into the prompt.

Review the attached product-image worksheet for [CATALOG]. Approved variants and stable item IDs: [LIST]. Authorized source photos: [FILE-TO-VARIANT MAP]. Flag duplicate IDs, wrong colors/items, mismatched image destinations and unresolved approvals. Separate observed mismatches from checks requiring the real public URLs. Return a correction list for the feed owner. Do not submit a feed, claim a business, invent an integration or promise product visibility.

4. Check the real URLs and complete the external handoff

Once the corrected assets are published by their authorized owner, open each direct image URL without relying on a signed-in editing session. Confirm that the image loads, shows the intended variant and remains available at a stable address. Open the product page too and check that its selected variant agrees with the row.

Have the feed owner validate the complete current required fields, countries, price, availability and submission rules against OpenAI’s documentation. Follow the actual onboarding route for the eligible business and preserve submission/validation outcomes. This demonstration made no external submission, so it supplies no acceptance or indexing result.

Choose an asset review before an integration project

Use this workflow when the catalog exists but images or variant assignments need checking. If the catalog facts themselves are unresolved, fix them first. If you need an automated feed integration, scope that separately with the authorized catalog and integration owners.

• Photoroom vs Ciyo product-photo workflow

Sources

Announcement and stable schema checked November 14, 2026. The River mugs example and suggested review method are original illustrative material.

• OpenAI: introducing Businesses in ChatGPT
• OpenAI: stable product feed schema
• OpenAI: submit product feeds