If ChatGPT Ads are not delivering, first specify the missing metric and reporting interval, then inspect the account, campaign, ad group and ad for explicit blockers and review state. Check constraints and reporting freshness next. GPTAd recommends changing one justified variable at a time; a bid increase cannot resolve every configuration, approval or measurement problem.
Reviewed October 15, 2026. This independent GPTAd checklist does not guarantee delivery, review timing, or a particular campaign outcome.
The quickest investigation usually begins by making the symptom more precise. Avoid changing bids, creative, and targeting together: even if delivery starts afterward, you will not know which change mattered.
Describe exactly what is missing
Choose the closest statement:
- There are no reported impressions.
- There are impressions but no reported clicks.
- There is activity but cost reporting looks incomplete.
- There are clicks but no measured business outcomes.
- There are measured events but no attributed conversions.
These are different investigation paths. Record the account and entity IDs, timezone, date range, extraction time, and the last deliberate configuration change. Capture a redacted screenshot or export before modifying anything.
For an agency, also confirm that the operator is viewing the intended client account. A technically valid investigation of the wrong account wastes time and can prompt harmful changes.
Walk down the hierarchy
OpenAI's troubleshooting guide recommends checking serving_issues and the account, campaign, ad-group, and ad hierarchy. Configured status and review state are distinct: active can coexist with pending review, while an approved ad can sit beneath a paused parent. An empty issue list does not guarantee delivery.
Make a four-row inspection sheet, one row per level. Record the current state, explicit blocker, relevant owner, and permitted next action. Read the whole sheet before editing a single row.
A hypothetical example: an ad appears approved, but its campaign is paused for an internal launch review. Raising an ad-group bid cannot resolve that condition. The right next step is to verify the launch decision with the campaign owner.
This exercise also prevents the opposite error: treating an intentional pause as a platform failure and undoing it without authorization.
Separate review from serving readiness
If review is pending, record that fact rather than inventing a deadline. If an issue requires a specific correction, address that issue narrowly and preserve evidence of the original response.
Creative changes can trigger review again, according to the same official guide. Repeatedly rewriting otherwise acceptable creative during an investigation may introduce a new waiting state instead of resolving the original problem.
Use a change log with the time, reason, owner, and expected observable effect. “Try something” is not a useful reason. “Correct the destination identified in the review response” is.
Inspect constraints after explicit blockers
Once the hierarchy has no unresolved explicit blocker, inspect the configuration as a combined system. Could several individually reasonable restrictions leave very few eligible opportunities? Is the intended offer actually available? Does the schedule match the reporting interval?
The troubleshooting documentation places targeting, opportunity volume, bids, and product availability among subsequent checks. It does not establish that a bid increase is the universal answer.
Write a hypothesis before making a change. For example: “The campaign is restricted to a smaller audience than the launch plan intended.” The evidence should be a configuration comparison, not merely low volume. If the restriction is intentional, ask the owner whether the experiment still serves its purpose rather than quietly broadening it.
Check reporting freshness separately
OpenAI notes that impressions and clicks may appear before finalized spend. For recent cost reporting, use returned CPC and CPM rather than rebuilding them from potentially incomplete spend. See the reporting guide.
A hypothetical dashboard showing activity with an unsettled cost field does not establish free traffic or a billing defect. Preserve the observation timestamp and compare a later snapshot. Do not convert a missing value into zero merely to make a chart render.
The appropriate next observation depends on the metric and its documented processing behavior. A few quiet minutes should not become a confident conclusion about campaign viability.
If delivery exists, follow the measurement path
Clicks without outcomes require a different investigation. Check the landing experience, business system, event producers, and selected reporting definition in sequence. The conversion setup guide treats event receipt, configuration, and campaign event attachment as separate steps; an incoming event alone does not prove attributed performance.
For escalation, provide entity IDs, exact times, redacted responses, the issue text, changes already attempted, and the specific expected behavior. Omit secrets and unnecessary customer data.
End with a classification: explicit configuration blocker, review state, constrained opportunity, reporting freshness, measurement defect, or unresolved issue. That classification gives the next owner something actionable. “Ads are broken” does not.
