Why Your ChatGPT Ads Conversions Number Can Move With No Campaign Change

A changed report is not automatically a changed campaign.

Processing lag and reporting settings are two reasons worth checking before touching a bid. They are not the only possible explanations. An event can arrive late, a customer can convert later, or the report can use a different definition. Those are different things to investigate.

The two real mechanisms

Processing lag. OpenAI tells advertisers to allow 24–48 hours for attributed conversions to appear, and that totals can keep changing while recent events finish processing (Measure Results). This is about data arriving late, not about eligibility rules shifting.

Reporting-setting differences. The same total can look different depending on: which time basis you're viewing (ad-event time, the default, vs. conversion time), your chosen click-through window (7/14/30 days), and whether the 1-day view-through window is on or off. Changing any of these changes what's displayed — it does not touch optimization, bidding, or billing.

Symptom	One explanation to check	Where to check
Yesterday's number keeps climbing today	Processing lag	Wait 24–48h before comparing
An export differs from an earlier export of the same period	A reporting window changed	Edit columns → Attribution windows
Two people report different numbers for the same date range	One is on ad-event time, the other on conv. time	Edit columns → Conversion Events
An expected conversion never appears	Eligibility, event configuration, or tracking failure	Compare event time, window, event name and received signals

Two hypothetical scenarios

Scenario 1 — processing lag only. A fictional campaign shows 40 attributed trials in a Monday-morning export. Six eligible Sunday conversions have not finished processing. A later export of the identical date range includes them and shows 46. In this constructed example, the customers had already converted; the report caught up. These numbers illustrate arithmetic, not typical volatility or a promise that every report settles after exactly 48 hours.

Scenario 2 — a settings change, not a performance change. In a separate fictional case, a teammate widens the click-through window from 7 to 30 days on Tuesday. Wednesday's report, for the same underlying user activity, now counts more conversions — because a longer window now captures clicks-to-conversion gaps that a 7-day window couldn't. No new customers converted; the ruler got longer.

An event outside an unchanged attribution window does not become eligible merely because another day passes. For example, a conversion eight days after a click remains outside a seven-day click-through window, unless the reporting rule or another qualifying ad interaction changes. Conversely, a new conversion occurring within the window can add to a report attributed to an earlier ad-interaction date.

Checklist before you conclude performance changed

1. Is the comparison inside the 24–48 hour processing window? If yes, wait.
2. Are both numbers on the same time basis (ad-event vs. conv. time)?
3. Did the click-through or view-through window change between the two dates you're comparing?
4. If comparing against your own analytics: same date range, same time zone, same event definition?

OpenAI states directly that a mismatch between its numbers and another analytics tool "does not necessarily indicate an error" (Conversion Measurement) — which is a reason to check settings first, not a reason to stop checking your tracking at all.

---

Method: Full text of two Help Center pages read September 19, 2026. Both hypothetical scenarios above are fabricated for illustration; no real campaign data or magnitude claim is implied.

Next action: Save the report settings beside each export. Run the checklist before interpreting a swing. Passing it makes the numbers more comparable; it does not by itself establish that a campaign change caused the result.