Three defects arrive together: an unpublished ad has a punctuation error, a live destination returns the wrong offer, and yesterday’s client report uses an incomplete period. Sorting that queue by discovery time would miss what matters. The defects affect different people, different decisions and different time horizons.
For ChatGPT campaign operations, a severity system should help someone choose the next action. It should not become a debate about whether a problem feels “major.” Define the consequences behind each label and keep the evidence close to the classification.
What is actually affected?
Write the observed defect in a single sentence before choosing its priority. “The destination linked from ad A shows an expired offer” is more useful than “landing page problem.” The first statement identifies what can be checked and suggests where the affected scope begins.
Distinguish an observed failure from a suspected one. A report discrepancy may come from different periods rather than corrupted data. An uncertain diagnosis can still need quick investigation, but uncertainty should remain visible. It is not a reason to invent a definitive cause.
Then identify the affected objects and people. Is this one draft, one active ad, every campaign using a shared destination, or an account-wide reporting transformation? A shared dependency can make an apparently small defect more important. Conversely, a dramatic-looking error in an unused draft may have no current customer exposure.
Use the current platform state as evidence. OpenAI’s campaign documentation distinguishes the state of ads and their parent objects. Do not infer actual serving solely from an ad’s creative preview. Read the relevant statuses and available serving information when deciding whether exposure is ongoing.
How does the defect change a decision?
A spelling error and an incorrect price can occupy the same number of characters but have very different implications. Ask what a customer, campaign operator or client reviewer could do differently because of the defect. That connection is a stronger basis for prioritization than visual prominence alone.
Consider a hypothetical monthly report with a mislabeled reporting period. If it is still an internal draft, the team may correct it through the normal review process. If a client is about to reallocate budget using it, the same defect has a more immediate decision impact. The underlying mistake has not changed; the operational context has.
For advertising copy, assess the meaning of the complete message. A missing qualifier may materially change an offer. A cosmetic spacing issue may not. Do not treat all text edits as low severity or all numeric differences as high severity without understanding what they imply.
Is it severe, urgent, or both?
Severity describes the consequence. Urgency describes how quickly a response is needed. A serious issue discovered before a scheduled launch may be contained while the launch remains paused. A modest report error may need prompt correction because a review meeting starts shortly.
The following is an example of an internal triage scheme, not an OpenAI classification or a universal service-level agreement:
| Classification | Example response logic |
|---|---|
| Current material exposure | Assign an incident owner and evaluate immediate containment |
| Decision at risk | Hold the affected decision or delivery while evidence is checked |
| Contained functional defect | Assign correction and verification before the next relevant release |
| Cosmetic issue | Schedule an editorial fix with a named owner |
Attach the actual scope and deadline to the label. “Decision at risk: Friday’s budget meeting, report section two” is actionable. “High priority” without context merely moves ambiguity into another field.
Do not bake unsupported financial thresholds into a universal rule. A spending difference that is small for one client may be material for another. Agree escalation thresholds with the responsible business owner and document their currency, period and scope.
Consequence and timing are separate questions
- Wrong live destination
Assess current customer impact and whether exposure needs containment.
Reconsider - Report before a budget meeting
Hold decisions that would otherwise rely on the defective evidence.
Reconsider - Functional defect in a paused draft
Correct and verify before the draft is used.
- Cosmetic text issue
Schedule correction after checking whether meaning is affected.
Who acts, and what closes the item?
Separate the person who owns the response from the person who performs a specialist correction. A destination error may need a website editor, while the campaign operator evaluates whether the affected ad should remain active. Clear ownership prevents each person from assuming the other has handled containment.
Choose a verification that matches the defect. A corrected destination needs the actual link and offer checked. A reporting error needs the relevant calculation and output reproduced. A screenshot saying a task is done is weak evidence if it does not show the property that was wrong.
Record classification changes as new evidence arrives. A supposed single-ad issue might affect a shared template. An apparent account-wide defect might be isolated to one export. Updating severity is responsible triage, provided the reason is documented rather than silently overwritten.
Connect the item to the team’s incident runbook. Use the more specific broken destination response when that is the failure, and review repeated patterns through a post-incident review. The outcome of classification should be a proportionate action with an owner and a testable completion condition. Check current platform state definitions in OpenAI campaign management.
Sources and scope
Assess a campaign defect by its scope, ongoing exposure and decision impact to choose an owner and the next response.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.