OpenAI now documents catalog replacement for eligible ChatGPT Ads product feeds. Treat it as a controlled inventory change, not routine file housekeeping: a partial export could remove products you intended to retain.
Documentation change observed on October 25, 2026, compared with October 24. This is not a confirmed launch date or a claim of general availability. GPTAd is independent of OpenAI; the safeguards below are our editorial recommendations.
How catalog replacement works
With is_full_refresh enabled, future full uploads replace the catalog; omitted products are deleted within a day. Delta updates are excluded. GET /v1/feeds/{feed_id}/settings reads the setting; PATCH changes it. Setting it to false stops pending cleanup but does not restore deleted products. Omitting the field leaves it unchanged.
OpenAI describes eligible feeds, without a country list or universal-access promise in this subsection. Confirm access for the intended feed before planning a production change.
Decide what a complete catalog means
Before enabling replacement, write a retention rule with the catalog owner: which products should remain represented, which are intentionally retired, and which are merely unavailable for now? A missing row should represent a business decision, not an export timeout.
For example, imagine a shop normally exports 12,000 product IDs but a failed supplier job produces only 8,000. This is a hypothetical quality-control scenario, not an observed ChatGPT Ads incident. An upload process that checks only whether a file exists could miss the missing supplier entirely.
Our recommendation is to compare the proposed export against the approved source inventory before transfer. Inspect added and removed IDs, variant relationships and counts by supplier or category. A plausible total alone can conceal one missing category and duplicates elsewhere.
Use an approval record, not just a toggle
Record the account and feed IDs, the current setting, the export version, the intended removals and the person approving them. Keep a copy of the last approved complete export and the proposed replacement in your own controlled storage.
Choose a business-specific stop condition for unexpected removals. Do not borrow a generic percentage as a safe threshold: removing one flagship product may matter more than removing many intentionally retired variants.
Separate a read-only settings check from authorization to change it. Agree who can enable replacement, who owns catalog completeness, and who can respond if the next export is incomplete. This article is not an instruction to enable it on every feed.
For the surrounding upload and tracking workflow, use our product-feed launch checklist.
Product-feed launch and tracking checklist
Plan the next upload and the recovery decision
Coordinate the change with whoever runs scheduled exports. Review whether another job could send an older or incomplete snapshot while the team is checking the first replacement. Keep the source inventory consistent across jobs rather than relying on whoever uploads last.
After an authorized upload, check processing status and diagnostics, including rejected rows, before comparing inventory with the approved list. A successful transfer does not establish completed processing. Investigate unexpected absences and preserve the export, timestamps and observations so a second reviewer can reconstruct what happened.
If an unexpected removal is suspected, escalate immediately to the feed owner. Distinguish stopping further cleanup from repairing the inventory already affected. Do not describe disabling the setting as an undo button, or promise a restoration time you have not verified.
If products are present but ads still do not deliver, investigate that separately rather than repeatedly replacing the catalog.
ChatGPT Ads delivery troubleshooting
The practical decision is whether your export process is trustworthy enough to make absence intentional. Keep replacement off until the responsible team has answered that question for its own catalog.
References
OpenAI: Product feeds, catalog replacement and feed operations
