ChatGPT Ads Troubleshooting: Statuses, Event Quality, and a Checking Order
Start From the Status Tooltip
OpenAI's troubleshooting guide starts every category the same way - read the status. An ad marked Not serving hides its reason behind a hover: common causes are ad copy or landing pages that conflict with the Ad Policies, or landing pages that block OpenAI's user agents or IP addresses. Policy problems resolve with an edit and a resubmit for review; blocked crawlers resolve on the landing page side, because a landing page the platform's agents cannot reach cannot be reviewed or verified. Account-level holds - verification, business verification, billing action - sit above the ad and show up the same way. Only when the tooltip gives nothing does the problem deserve the deeper checks below.
Setup and Launch Failures: Verification, Billing, Uploads
Three setup clusters cover most launch-day failures. Access: sign in with the OpenAI account tied to the right organization, and if the Ads Admin role is missing, an IT admin grants it in the OpenAI admin console through the user's direct roles. Billing: campaigns will not deliver until a billing profile and a payment method exist - the setup requirements from the campaign setup guide. Uploads: a campaign file that errors should be compared against OpenAI's campaign structure checklist before re-upload, and images have hard requirements - PNG or JPG, square, no larger than 1200 by 1200, publicly reachable at a direct image URL (a direct download, not a viewer page) from hosting such as Google Drive links, AWS buckets or CDNs, or the advertiser's own domain.
No Impressions: The Pre-Delivery Checklist
A launched campaign with zero impressions has a short official checklist. Confirm billing is fully set up and the payment method is valid, the campaign budget has not been reached, the campaign end date has not passed, and account verification completed successfully. Actively delivering ads should show a Serving status - if that status is present but numbers stay at zero, the delay is the next suspect rather than the setup: OpenAI asks for at least 24 hours after launch before flagging delivery issues, and there is typically a delay of up to 7 hours between ad delivery and its appearance in Ads Manager reporting. Waiting is genuinely part of the attribution mechanics, not a support dodge.
Event Quality Warnings: What They Mean and What to Fix
Event quality is OpenAI's per-data-source assessment of whether conversion information is useful for matching, currently in beta and refreshing daily using seven complete calendar days - which means improvements appear gradually as older events leave the window. The score and its warnings highlight setup areas to review; they do not predict campaign performance or guarantee that any conversion will match to an ad. The warnings translate directly into fixes. Limited email or customer ID information asks for an email or a stable customer identifier on eligible conversions, normalized and hashed per the integration instructions, with the same identifier reused for the same person. Ad click information is missing points at the oppref chain: preserve the parameter through redirects and pass click information into server events. Delayed Conversions API events have a concrete bar - OpenAI checks for receipt within one hour of the recorded action, and retried events keep their original event time rather than the upload time. Conversion matches need review means some potential matches were removed during attribution checks - usually repeated callbacks for one action, inconsistent timing, or too many matches tied to one ad view - and the fix runs through dedup: a consistent event_id for browser and server copies of the same action, distinct IDs for distinct actions, and the original timestamp preserved. One rule sits over all of them: use real customer information and real business actions - extra events, invented identifiers, and shifted event times to raise a score are explicitly out.
When Numbers Look Wrong
Reporting surprises usually trace to comparison hygiene. OpenAI's guidance: confirm the ads in scope actually show Serving, and when comparing across table views, CSV exports, and charts, apply the same campaigns, filters, and date range in each. Conversion counts follow their own clock - the standard 24-48 hour inclusion delay applies, which the attribution guide breaks down together with the differences against tracker numbers. Only after those two checks does a discrepancy become a support case, and the troubleshooting guide asks for the evidence to travel with it: the export, the date range, the filters, and a note on what was expected.
A Debugging Order That Works
The order matters because each step eliminates a class of causes. Read the status tooltip. Then walk setup: access and roles, billing profile and payment method, file and image requirements. Then delivery: the Serving status plus the 24-hour patience window and the 7-hour reporting lag. Then measurement: event quality warnings and the conversion tracking validation flow, with the pixel's debug mode logging SDK activity in the browser console while you test. Support comes last, and it arrives prepared - OpenAI documents which details each issue needs, from error screenshots to image URLs, so the ticket resolves in one pass. MOST delivers tracked conversions server-side to ChatGPT Ads alongside Meta, TikTok, Reddit, Snapchat, and Pinterest, and the free Pixel Activator validates one event per landing at pixel.way2.us so measurement problems never reach the campaign stage at all.