Google Merchant Center to OpenAI: Feed Mapping

Write a mapping contract before converting rows

For every target field, record the source column, any transformation and the behavior when data is absent. Distinguish a direct copy from a merchant-provided default and from an approved descriptive change. This small contract makes future refreshes easier to explain.

Start with a few real variants. Check that IDs remain text, links identify the correct selection and images show the same item. Preserve the source row in the review record so a downstream finding can be traced back without reconstructing a spreadsheet by hand.

Resolve naming differences explicitly

OpenAI’s Stable discovery schema uses its own field names and value conventions. A Merchant Center-style source can provide much of the underlying information, but a value that is familiar in one source may need a deliberate target mapping. Availability is a common place to check carefully.

Treat seller context and brand as separate facts. The store operating the feed may not be the brand of every item it sells. Do not use a convenient hostname or a generic placeholder to satisfy a required field without confirming that it represents the actual seller or product.

Keep transformations narrow

Normalize whitespace or serialize a value where the target contract requires it, but do not silently change the payable amount, selected variant or meaning of a description. A mapping preview should make these transformations visible before a release is prepared.

For price and currency, use the actual source offer and the target’s money representation. For optional descriptive fields, prefer supported facts over guesses. A missing value may require a supplier request or merchant default rather than a more aggressive conversion rule.

Test the derivative independently

Validate the proposed output against the current target requirements and compare it with the original source. Include missing values, duplicate IDs, sale cases and variant groups. Keep a record of the source snapshot, mapping version and findings.

Finally, separate output preparation from delivery and acceptance. An exported artifact can be useful while platform onboarding is still pending. Make that status visible to everyone evaluating the workflow, particularly when a commercial proposal uses broad language such as ready for ChatGPT.

Merchant Center to stable OpenAI field mapping

The following table is a starting map for the Stable file-upload reference checked on 31 October 2026. Similar field names do not remove the need to inspect the values. Keep your source column, target field, transformation and evidence in the mapping record so another operator can reproduce the result.

Seller context deserves a separate decision. A multi-brand shop can sell a product made by a different company. Confirm the seller once at the appropriate merchant scope, then inspect product-level exceptions. A convenient store name should not overwrite the product’s brand.

{
  "item_id": "000042-BLK-M",
  "title": "Harbor tee \u2014 black, medium",
  "description": "Cotton jersey tee. Color: black. Size: medium.",
  "url": "https://example.com/tee?variant=black-medium",
  "brand": "Harbor Loom",
  "seller_name": "Example Outfitters",
  "image_url": "https://example.com/images/tee-black.jpg",
  "availability": "in_stock",
  "price": "24.00 USD"
}

Test transformations with records that can fail

Use a small mapping test set with expected results written before the transformation runs. Include an ID with leading zeros, a description containing a comma or newline, a pre-order item and two variants with different stock states. Check both the parsed values and the serialized file; a correct value can still be damaged by incorrect CSV quoting or numeric coercion.

Then compare the release with the public product pages and your source of stock and price. If the site and feed disagree, establish which source is current before changing either. Preserve the previous mapping and release so a correction can be traced. Re-run the same cases when the source headers or target schema change.

Your practical checklist

1. Document copy, default and transformation rules.
2. Preserve string IDs and variant selection.
3. Confirm brand and seller independently.
4. Preview changed commercial values.
5. Keep delivery and acceptance outside the local validation result.