Illustrative photograph. Photo: Shopping cart with a single apple on a wooden table indoors.jpg by Shixart1985, via Wikimedia Commons — CC BY 2.0. Downloaded Wikimedia version; no custom edits.
A shopper asks for a waterproof jacket in a specific size under a fixed budget. Your catalog contains the jacket, but the product title hides its intended use, the image shows a different variant, and the listed price belongs to the smallest size.
More distribution would spread the inconsistency. Start by checking the product information you already publish.
Separate the feed from the rumored profile
OpenAI documents product feeds for catalog discovery. Its onboarding guide currently limits access to approved partners. That documented merchant path is separate from reports about a broader business hub; it is not evidence that every business can claim a universal profile. See feed onboarding.
Preparing your catalog can be useful before eligibility is established. Present an integration as live only after your own access and delivery have been verified.
Audit a sample before the whole catalog
Choose a small sample containing a bestseller, a product with multiple variants, an out-of-stock item, a discounted item, and a product with a common compatibility question.
For each, compare the source catalog, public product page, cart behavior, and any active distribution feed.
Check | Question to answer |
|---|---|
Identity | Does this record refer to one identifiable item or variant? |
Price | Does the displayed value match the selected item and currency? |
Availability | Can a shopper actually buy it now? |
Images | Do they show the selected variant accurately? |
Destination | Does the link open the expected product and selection? |
Conditions | Are shipping, compatibility, and exclusions understandable? |
Treat discrepancies as defects with owners, rather than as a copywriting task alone.
Model variants carefully
Check a fictional shoe offered in black and white, with several sizes. A black size-eight record should not lead to an unrelated default selection or use an image that implies a different product.
Use stable identifiers and approved attribute values. Do not invent a barcode or certification to make a record look complete.
The current OpenAI product schema distinguishes supported Stable fields from Draft fields, and notes that eligibility does not guarantee display. Follow the production contract for your actual integration. Read the product schema.
Make product descriptions answer buying questions
Replace unsupported superlatives with useful specifics. Explain intended use, materials, dimensions, compatibility, care, and meaningful limitations where relevant.
For a laptop accessory, the important question may be which models and ports it supports. For a jacket, it may be the meaning and evidence behind a weather-resistance claim.
Use approved product evidence. Do not infer technical certifications from a supplier’s promotional language.
Assign an owner to freshness
Decide which system owns price and stock, how updates reach each destination, and what happens when delivery fails. A feed that uploaded successfully last week may still be wrong today.
Track rejected records, stale updates, broken destination links, and variant mismatches. Review resulting purchases alongside returns and customer complaints where your measurement permits it.
The first success criterion is simple: the item a customer discovers should be the item they can buy, under the terms they were shown.
Next read: Measure commercial outcomes separately from visibility.
