01 Access and authorization
Complete the merchant application and get the feed access method and credentials the platform confirms. A model API key does not grant product feed access.
PLATFORM PLAYBOOK / 01
ChatGPT Product Discovery
From your catalogto the OpenAIproduct feed.
See how fields map, how the feed is delivered and how price and inventory stay current — planned against OpenAI's Stable spec.
OpenAI confirms feed access per merchant. This page describes a service; it is not a self-serve submission tool.
CATALOG → DISCOVERYIllustration
Start from a shopper question
“A storage cabinet for a small space, under 60 cm wide?”
01
Stable identitySKU / variant
02
Product factsSize · material · use
03
Transaction dataPrice · stock · product page
Match shopper needs to product facts you can back up.
Reuse your product masterSourced AI enrichmentMapped to each platform's specTraceable updates and releases
PRODUCT FEED INTEGRATION
How does your catalog connect to OpenAI?
Read the source → map fields → validate against the spec → submit → read results back. This is the integration design, not a live connector.
02 Export and submit
The file route pushes full snapshots over SFTP in Parquet, jsonl.gz, csv.gz or tsv.gz; production integrations use the Stable spec. A Google-compatible format is a separate route to confirm.
03 Updates and removals
OpenAI recommends refreshing full snapshots at least daily under a stable file name. Products left out of a snapshot aren't removed immediately and can remain for up to 14 days; send is_eligible_search=false to stop search eligibility and let the platform process it.
Field mapping: get the spec right first.
The left column is a source-data concept, not a verified field name in your system; required, conditional and market rules follow the official spec.
| Source data concept | OpenAI Stable field | Mapping and validation |
|---|---|---|
| SKU / variant ID | item_id | One row per purchasable variant; the ID doesn't change with price. |
| Product name / description | title / description | The description must match the product facts. |
| Product page / main image | url / image_url | Link to the right variant; images must be publicly readable. |
| Brand / seller name | brand / seller_name | Map them separately; brand alone isn't enough. |
| Price / stock | price / availability | Price includes currency; stock maps to the spec's enumerated values. |
| Parent product / options | group_id / listing_has_variations / variant_dict | Provide when variants exist; never merge them into one SKU. |
An integration you can accept and audit.
Traceable batches
Keep product IDs, source versions, submission times and results. Catch missing required values first; retry transport failures, but don't resend rejected data forever.
Fix, then verify again
Format errors fix the mapping; data conflicts go back to the source; permission and policy issues go to their owners. Inclusion and recommendation are never guaranteed.
Integration docs: OpenAI file delivery spec ↗
KNOW THE BOUNDARIES
Confirm the access route before you publish.
Being clear about what each platform supports, and what onboarding needs, comes first.
Eligibility
Merchants apply and confirm a supported integration. If your e-commerce platform already sends your data, check that route first so you don't create a duplicate source.
Markets
OpenAI's merchant page currently describes the shopping experience for users in the United States. Your location, shipping markets and eligibility still need confirming case by case.
Discovery ≠ checkout
The product feed powers discovery and how products are shown. Feed submission, ad feeds and checkout integration are separate eligibilities.
DATA, NOT GUESSWORK
Every field has a job to do.
These are onboarding checkpoints, not the full platform spec; required fields depend on category, market and the approved format.
| Area | Data | How we plan it |
|---|---|---|
| Identity and variants | Product ID, brand, product / variant URL | Keep IDs stable so sizes, colors and pages never mismatch. |
| Content and evidence | Title, description, images, verifiable attributes | Take sizes from the product spec; claims without evidence go back to the brand. |
| Price and availability | Price, currency, inventory | The source system is the truth; re-export after changes and never let AI rewrite transaction facts. |
| Output format | Fields mapped to the Stable spec and an approved format | Draft specs are for planning only. A Google-compatible route also needs confirming, and XML isn't accepted. |
WORKED EXAMPLE
From one question
to one fix you can make.
An illustration, not customer data or a platform report.
- Spot
The description only says “minimal storage cabinet”
- Check
The official spec page lists a width of 55 cm
- Fix
Add the size and unit, keep the source, wait for brand sign-off
FROM CATALOG TO MARKET
Don't rebuild your catalog. Build a publishing flow.
Your product system keeps the original facts; AI Catalog manages enrichment, review history and platform outputs.
- 01
Confirm eligibility
Check markets, existing routes and how OpenAI approves you.
- 02
Enrich the data
Map fields, check gaps, get brand review.
- 03
Prepare submission
Export to the approved spec and keep batches and versions.
- 04
Keep it current
Sync price, stock and delistings; track whatever processing status is available.
A successful submission isn't the finish line.
Publishing health
Track available processing status, data freshness and open issues.
Observable results
Watch exposure and referral traffic separately; an accepted product is not proof of an AI recommendation.
BEFORE YOU START
Worth asking first.
No shortcut guarantees ranking.
A data flow you can check and fix does help.
Can I upload products to ChatGPT shopping with an OpenAI API key?
No. Sending data to a model API is not the same as joining product discovery. The product feed goes through OpenAI's merchant application and the integration it approves.
Can my existing feed be sent as it is?
Don't assume so. Identity, fields, format, target market and eligibility all need checking. AI Catalog keeps your product master and builds an output mapping for each destination.
Will complete data guarantee ChatGPT recommends my products?
No. Submission, processing, inclusion and recommendation are different states. Complete data helps products show accurately but doesn't guarantee exposure or ranking.
Can I submit directly on this site?
No. This page describes the service and onboarding. The homepage scan is a readiness check; integration, sync and monitoring scope are agreed in a pilot.
START WITH YOUR CATALOG
Start with your product data.
Bring your target markets, catalog scope and platform account status, and we'll scope a pilot.
Book a catalog consultation Opens Crescendo Lab's official contact page; this page collects no data and connects to no platform.