Half your catalog went missing on the way to ChatGPT, and the upload report shows your OpenAI product feed rejected rows you thought were fine. OpenAI’s product feed specification is strict about a few values. On stock status it says Omitted, empty, or unrecognized values reject the row.
(OpenAI’s product feed specification).
In short: OpenAI judges each row on its own. A row fails when a required field is empty or a value isn’t in OpenAI’s format. In the cases we’ve read, most failures come from a feed built for Google Merchant Center and sent to OpenAI with a few names and values unchanged. Here is what OpenAI asks for, where a Google feed differs, and how to fix the rows.
Why was my OpenAI product feed rejected?
Usually it wasn’t the whole feed. OpenAI processes one row per item or variant, and rows fail one at a time.
OpenAI lists nine required fields for every row: item_id, title, description, url, brand, seller_name, image_url, availability and price. The spec says Required fields must contain a value
. OpenAI’s file upload guide names the usual reasons a feed fails to load:
-
Missing required fields
-
Outdated or non-spec field names
-
Malformed field values
The availability rule is the only one the spec spells out as a row rejection. For other fields, OpenAI also warns that its format notes don’t mean every bad value is caught at upload. Our reading: a value can pass the upload and still be wrong in ChatGPT, so fix formats even where nothing was rejected.
If you upload through ChatGPT Ads Manager, the Ads product feed guide shows each upload’s accepted and rejected row counts, with diagnostics by field. For an invalid_value diagnostic it says: Check the reported field’s formatting and allowed values.
Which availability values does OpenAI accept?
Five, written exactly like this: in_stock, out_of_stock, pre_order, backorder and unknown. Anything else rejects the row. OpenAI adds: Use unknown explicitly when stock status is unavailable.
This is where reused feeds break. Each system has its own spelling:
| Where the value comes from | In stock | Out of stock | Pre-order | Backorder |
|---|---|---|---|---|
| OpenAI accepts | in_stock | out_of_stock | pre_order | backorder |
| Google Merchant Center | in_stock | out_of_stock | preorder | backorder |
| WooCommerce stock status | instock | outofstock | (none in core) | onbackorder |
| Product structured data | InStock | OutOfStock | PreOrder | BackOrder |
Google’s values come from its availability attribute page. WooCommerce’s come from its REST API documentation. Two traps stand out:
- Pre-order. Google writes
preorder. OpenAI writespre_order. A Google feed sends every pre-order row with a value OpenAI doesn’t list. - Raw platform values. A feed that copies WooCommerce’s
instockor a page’sInStocksends values OpenAI doesn’t recognize either.
Google also asks for an availability date with pre-order and backorder items. OpenAI won’t act on one: Sale-window dates, expiration dates, and availability dates do not schedule price or availability changes in this contract.
Send the current status in every snapshot instead.
The value must also be true. Google says it plainly: The values in your data source should always match the information on your website.
Our availability mismatch guide covers how stock drifts between a store and its feed.
Why does a Google Merchant Center feed fail in OpenAI?
The two formats look alike, so it’s tempting to reuse the Google one. These are the differences that matter:
| Field | Google Merchant Center | OpenAI |
|---|---|---|
| Product ID | id | item_id (id still accepted) |
| Variant group | item_group_id | group_id (item_group_id still accepted) |
| Product page | link | url |
| Main image | image_link | image_url |
| Seller | no such field | seller_name, required on every row |
| Brand | required for most new products | required on every row |
| Pre-order | preorder | pre_order |
OpenAI keeps a short list of older names it still accepts, including id and item_group_id. Google’s link and image_link aren’t on it. Our reading: a Google feed with its own column names arrives without a url or image_url, which are required.
Brand gets lost too. OpenAI’s spec describes brand as Product brand as shown on the product page.
It adds: Use a real brand and seller name, not placeholders.
If your Google feed skips brand on some products, OpenAI’s will too. Our missing brand guide shows where brand lives on each platform.
Price is the one field where the two formats agree. Google’s price attribute page uses 15.00 USD, and OpenAI uses 79.99 USD. Price rows fail when the feed tool copies the price as your store displays it, with a comma for thousands or for decimals. OpenAI’s rule: Use a decimal point, no thousands separators or exponent notation, and no more fractional digits than the currency permits.
Google’s rule on the number itself applies here too: Submit an amount and currency that match the price on your landing page and the checkout pages.
Last, the market. The standard OpenAI-format upload currently targets the US.
OpenAI says other markets need its confirmation first, so a store that prices only in another currency should ask its OpenAI contact before sending a feed.
How do I fix rejected rows?
- Find the failing field. In Ads Manager, open the upload history and read the diagnostics: each one names a field and how many rows it hit. OpenAI’s file upload guide doesn’t describe a rejection report for discovery feeds, so if you send one, ask your OpenAI contact where to see it.
- Check the header row. All nine required columns, in OpenAI’s names:
item_id,title,description,url,brand,seller_name,image_url,availability,price. - Map availability. Translate every stock value to one of OpenAI’s five. Map pre-orders to
pre_order, notpreorder. - Fix the price format. Amount, a space, then the currency code:
1299.00 USD, not1,299.00 USDor1299,00 USD. - Fill brand and seller name. Use real names on every row. Don’t send
n/a,nullorunknownin a field where OpenAI doesn’t list it. - Keep identifiers as text. OpenAI asks you to keep identifiers as strings so leading zeros survive. Opening a feed in a spreadsheet can drop them from GTINs.
- Test small, then go daily. OpenAI’s advice:
Start with a small sample (around 100 items).
Then send a full snapshot at least daily under the same file name. - Check the values against your store. A row that loads with the wrong price or stock is worse than a rejected one. Compare a few rows with the live product pages.
On Shopify, WooCommerce and other platforms
Shopify
On Shopify you usually don’t build an OpenAI feed. Shopify’s ChatGPT channel page says the channel is active by default for eligible stores, and products reach ChatGPT through Shopify Catalog. Its Catalog page says Shopify Catalog is the authoritative product data feed to your agentic channels.
So the job moves upstream: if a product is wrong in Shopify, it’s wrong in Catalog. Check Sales channels > Agentic for your Catalog count or an Ineligible warning. Our ChatGPT shopping post for Shopify covers eligibility. The Shopify pages we read don’t mention ChatGPT Ads feeds.
WooCommerce
WooCommerce stores usually build the OpenAI feed with a feed plugin. Check its field mapping, not just its Google settings: the stock status mapping (instock must become in_stock), the price format, brand, and a seller_name set to your store’s name. Regenerate the feed after each mapping change, and clear any page cache so the feed and the pages show the same stock.
Other platforms
Whatever builds your file, OpenAI’s spec is the reference. Use UTF-8, absolute HTTPS URLs that anyone can open, and one row per variant with its own price, stock, URL and image. OpenAI says Product and image URLs must be publicly accessible.
What we see in real cases
Paraphrased from public merchant forums, with no names or stores.
- A WooCommerce store selling outside the US sent a ChatGPT feed and saw only about half its rows accepted. The fixes were the ones above: availability values corrected, thousands commas taken out of prices, descriptions trimmed.
- A WooCommerce merchant found brand missing from every product in the OpenAI feed file, while the Google and Meta feeds from the same plugin had it.
- A feed vendor found that a hosted CSV link failed in ChatGPT Ads Manager while a TSV file and a manual upload worked.
- An advertiser had every product in a small ChatGPT Ads feed marked as not passing policy, with no rule named and no appeal button. That is a policy decision, not a format rejection, and fixing columns won’t change it.
- Applicants to OpenAI’s merchant form reported waiting weeks with no confirmation. OpenAI’s Get Started page says
Onboarding product feeds in ChatGPT is currently available to approved partners.
The pattern: a feed fails on a handful of values, and the same few values fail on every row. Fix the mapping once and most rows come back.
Accepted isn’t the same as shown. OpenAI’s spec says search eligibility doesn’t mean display, and its Ads guide says A completed upload does not establish that every product can serve.
Google only shows your products if it trusts your store, and the product pages behind both feeds are the same. The free store check doesn’t read OpenAI feeds; it reads your product and policy pages as a visitor sees them, and Price Watch compares each page’s structured-data price with the price the store shows.
What OpenAI doesn’t say
- Whether a row with a malformed price, GTIN or URL is rejected, or loaded and ignored.
- Whether field names and values are matched regardless of capitalization. Send them exactly as the spec writes them.
- How long a corrected row takes to show in ChatGPT, or whether it will.
FAQ
Which availability values does OpenAI accept?
OpenAI’s spec lists in_stock, out_of_stock, pre_order, backorder and unknown. It says omitted, empty or unrecognized values reject the row, so Google’s preorder, WooCommerce’s instock and schema.org’s InStock all need mapping first.
Can I upload my Google Merchant Center feed to OpenAI as it is?
No. Several Google attribute names, such as link and image_link, aren’t on OpenAI’s list of accepted names, Google has no seller_name field, and Google’s preorder value isn’t one of OpenAI’s. Map the columns and values first.
What price format does OpenAI’s product feed need?
An amount in major units, a space and a three-letter currency code, such as 79.99 USD. OpenAI asks for a decimal point and no thousands separators.
If all my rows are accepted, will my products show in ChatGPT?
Not necessarily. OpenAI’s spec says eligibility doesn’t mean display, and its Ads guide says a completed upload does not establish that every product can serve.
Do Shopify stores need to send OpenAI a product feed?
Shopify’s help center says ChatGPT gets your products through Shopify Catalog and you don’t need to do anything for them to be discovered there. Shopify’s pages don’t describe building a separate OpenAI feed.