OpenAI Stable vs Google-Compatible Product Feeds

Checked against OpenAI's published product-feed reference on 29 September 2026. This guide focuses on mistakes that matter when a merchant or agency reuses catalog data across the two profiles.

This is an independent technical guide, not OpenAI or Google certification. Platform specifications can change; verify current official documentation before production submission.

The important difference

Do not treat the two profiles as interchangeable CSV layouts. OpenAI's Stable discovery format requires nine basic fields per purchasable item: item_id, title, description, url, brand, seller_name, image_url, availability, and price. The same reference explicitly distinguishes Google-compatible feeds and merchant identity handling.

Issue	OpenAI Stable	Google-compatible implication
Seller identity	`seller_name` is required on each row.	OpenAI's reference says Google-compatible feeds use the registered merchant name instead; do not assume a row-level seller field maps directly.
Availability	`in_stock`, `out_of_stock`, `pre_order`, `backorder`, or `unknown`.	A converter should not silently turn an unknown stock state into an affirmative stock state. Preserve uncertainty for manual review when the target profile cannot represent it safely.
Price	Money is written as amount + ISO currency, for example `79.99 USD`.	Validate the target profile separately and keep price consistent with the landing page; do not infer or fabricate a payable amount.
Missing facts	Unknown optional values should generally be omitted rather than filled with placeholders, except where a value such as `unknown` is explicitly allowed.	Conversion should not invent brand, identifiers, stock, seller identity, or other merchant facts just to make a file pass.

A safer conversion workflow

Audit the source feed before conversion.
Separate deterministic formatting fixes from missing business facts.
Keep a changelog of every automatic edit.
Block or flag values that cannot be mapped without changing their meaning.
Validate the converted file against the target profile again.
Check the public product page for canonical/indexing signals and Product structured data where relevant.

Example: why `unknown` matters

If an OpenAI Stable row says availability is unknown, replacing it with out_of_stock or in_stock changes the merchant's factual claim. A safe converter should leave the target value unresolved and require review instead of inventing inventory status.

Where the toolkit fits

The AI Shopping Feed & Client Audit Toolkit for Windows is built around that preflight workflow: local CSV/TSV auditing, separate OpenAI and Google-oriented checks, conservative safe fixes, change evidence, rechecks and client bundles. It does not submit feeds or call platform APIs.

See the public sample workflow →

Get the Windows toolkit — $19 one-time →

Primary source