Create a discovery-data draft
ChatGPT Product Feed Generator Guide
A generator is useful when you need to map an existing catalog export into a structured discovery-data draft without confusing that draft with a live integration.
Validate a feedWhen a feed generator helps
A generator can turn a spreadsheet or catalog export into a consistent draft with one purchasable item or variant per row. That is valuable when the source data uses different column names or separates variant information across systems.
The useful output is a reviewable draft, not a promise that a platform will accept, publish, or surface the products.
Start with a minimal field map
Map source columns to the nine discovery fields before adding optional attributes. This keeps the first pass focused on product identity, buyer-facing detail, URLs, seller identity, price, and availability.
If a source field is unknown, leave it as an explicit data gap for a merchandiser or catalog owner to resolve. Do not fill it with placeholder text.
- Keep one stable item_id for each purchasable item or variant
- Use public absolute HTTP or HTTPS product and image URLs
- Represent price as an amount with a three-letter currency
- Use an allowed availability value from the current specification
Treat variants as real products
A variant with a different price, availability, URL, or image needs its own accurate product data. Do not reuse parent details when they would make a shopper see the wrong offer.
The official product-feed reference explains parent and variant identifiers in more detail. Use it before deciding how to model a complex catalog.
Validate before delivery
Catalog Signal can normalize an existing local file and surface basic gaps. It does not currently create a feed from a Shopify connection, host a file, schedule delivery, or request platform access.
After local validation, compare the draft with the current official specification and the delivery path available to your merchant setup.
Common questions
FAQ
Should I use JSONL or CSV?
Choose the format supported by your confirmed delivery path. OpenAI's current documentation is the authority for supported formats and onboarding details.
Does a generated draft upload products to ChatGPT?
No. A draft is just structured data. Upload, eligibility, and product display require a separate approved platform workflow.
Do Ads and checkout use the same requirements?
No. OpenAI documents discovery, Ads, and checkout as separate scopes with additional requirements for some workflows.