All articles

ChatGPT Ads Feed Reviews: Separate Pending, Rejected and Unknown

Newly documented feed-review diagnostics and a revenue-reporting caveat call for clearer operational queues, access checks and honest treatment of missing data.

Original illustration of connected pastel cards with nodes, a key and a padlock.

Keep pending reviews, rejections and missing information separate in your ChatGPT Ads dashboard. OpenAI’s updated feed documentation gives eligible accounts more diagnostic detail; a reporting clarification also explains why some purchase counts have no revenue value.

Documentation changes observed November 18, 2026, against our November 17 check. This is not a confirmed launch date or a promise of access for every account. GPTAd reviewed documentation, not a live advertiser integration.

What changed in feed reviews

The Product feed guide now describes approval-list review statuses and optional rejection explanations. A false approval flag can include pending items; a missing explanation does not establish approval. Access is account-dependent.

Hotel feeds remain a limited, approved-advertiser beta. Their item-query response can expose review status and optional reasons for accounts with review access. Inventory eligibility and policy approval are separate.

Build a decision queue, not a red-number dashboard

Our recommendation: give each diagnostic record an account, feed, returned identifier, observation time and assigned owner. Preserve the raw result in a restricted operational log, then present a short human-readable action beside it. Do not turn every unknown into a failure notification.

Use three separate work queues: waiting for a decision; confirmed rejection requiring review of the stated issue; and incomplete evidence needing an access or data check. Choose the queue only from evidence you actually have. Keep these labels distinct from your business decision to pause or continue a campaign.

For a hypothetical retailer, a merchandising owner might inspect the product page while an engineer checks the lookup response. Record what each person verified before re-uploading anything. This avoids changing a valid catalog merely because a dashboard used an oversimplified label.

Add regression tests around unknown states and stale observations. A useful acceptance test is whether a colleague can explain why an item appears in its queue without reading the implementation. Include an explicit “not yet verified” outcome rather than forcing a green or red result.

A separate revenue-reporting caveat

In ad-level POST /v1/conversions/insights reports, conversion_time can count goal-matching purchases while order_created_attributed_sales is null. OpenAI directs advertisers to ad_event_time for daily counts and revenue together.

For your dashboard, retain a visible distinction between zero and unavailable. Put the reporting basis beside the chart and agree it before comparing periods. Do not silently fill an unavailable monetary value with zero or interpret it as proof that purchases had no value.

Continue with our feed launch checklist.

Use our attribution reconciliation workflow before comparing reports.

References

OpenAI: Product Feeds

OpenAI: Hotel property feeds, limited beta

OpenAI: Reporting

Good questions open new possibilities.
Put these ideas to work

Your next step.
Handled together.

Talk to our team about SEO, AI discovery and ChatGPT advertising, or get early access to the workspace behind our work.

Where we can help
  • Technical SEO and content
  • AI discovery on ChatGPT, Perplexity and Gemini
  • ChatGPT ads, when they fit