A person interacts with a ChatGPT ad on 29 September and purchases on 3 October. Should the conversion appear in September or October? Both can be valid reporting answers, depending on whether the report groups outcomes by the credited ad interaction or by the conversion date.
This is a time-basis choice. It is different from a time-zone choice, which determines the calendar date of a particular instant. Moving between Stockholm and another zone does not resolve whether the report should follow an earlier ad interaction or a later purchase. First choose the event that anchors the date, then apply the appropriate clock.
Two views of the same outcome
OpenAI’s dedicated conversion endpoint defaults to attribution_time_basis: "ad_event_time", which dates outcomes by the credited advertising interaction. Choosing "conversion_time" dates them by the outcome instead. Coverage of non-goal events is limited under conversion time; use ad-event time when examining those events. Otherwise an apparent disappearance can reflect the chosen report view rather than failed event delivery. These behaviors are described in OpenAI reporting.
For the hypothetical September interaction and October purchase, an ad-event view can assign credit to the September interaction period. A conversion-time view can place the outcome in October. This does not create two purchases. It organizes the same attributed outcome under different reporting questions.
Label the date column explicitly. “Date” is insufficient in a report where both views are used. “Attributed interaction date” and “Conversion date” let the reader understand why a historical period can change after later purchases occur.
Which event determines the month?
- Ad-event time
The outcome follows the credited September interaction.
- Conversion time
The outcome is organized around the October purchase.
- One business event
Two date views do not create two purchases.
Choose the question before the view
If the decision concerns what a defined set of September advertising interactions produced, an interaction-based view can be appropriate. It supports evaluating the outcomes associated with that acquisition period, subject to the reporting rules and outcome maturity.
If the question concerns when attributed outcomes occurred operationally, a conversion-time view can be useful. It may help compare the timing of attributed purchases with business activity, but it still does not become a complete store-sales ledger. It remains an attribution report with its own scope.
An executive report may need both, shown separately. The campaign analyst can review acquisition cohorts while an operations reader sees when outcomes arrived. Combining the two into one unlabeled monthly total would lose the distinction that makes each useful.
Avoid the calendar CPA trap
Suppose, hypothetically, September advertising spend is 2,000. Some credited purchases happen in September and others in October. Dividing September spend by only September conversion-date outcomes does not necessarily measure what September’s advertising produced.
That calculation can still be a defined calendar-period ratio, but its meaning is narrower than cohort acquisition cost. Name it accordingly and avoid comparing it directly with an interaction-cohort CPA. The denominator may include outcomes from earlier advertising and exclude outcomes that have not occurred yet.
For a cohort analysis, ensure the cost and outcome population correspond under a coherent method. If the available data cannot establish the linkage, keep the figures separate. A formula should not manufacture the relationship merely because spend and conversion count share a calendar-month label.
Expect historical movement under interaction time
When later outcomes are assigned to earlier interactions, the earlier period can gain conversions after the calendar interval has ended. That is not automatically a data error. Preserve the retrieval timestamp and define when the business considers a report sufficiently mature for its decision.
Do not promise that a chosen monthly closing date makes the historical figure permanently final. Reporting updates and business adjustments can still matter. The month-end close process should explain how later changes are handled rather than pretending they cannot happen.
If the organization changes its primary time basis, annotate the change and regenerate comparable history where possible. A trend that switches basis halfway through can show a jump unrelated to campaign performance. The old and new views need an explicit bridge.
Test with a boundary case
Before using a new reporting integration, choose a safely controlled example that crosses the reporting boundary. Verify how the relevant outcome appears under each supported basis and record the request settings. Use permitted test procedures and avoid creating misleading live business events merely to populate a report.
If individual event evidence is unavailable, validate using a clearly scoped aggregate comparison and document what it cannot prove. Do not infer user-level journeys from totals. The objective is to confirm the report convention, not to claim access to personal conversations or behavior.
Use time-zone alignment for day boundaries and mature click cohorts for equal observation time. The platform source is OpenAI reporting. A well-labelled ChatGPT report answers one timing question at a time and prevents a date convention from masquerading as a change in advertising effectiveness.
Sources and scope
Choose whether conversion reporting dates refer to the credited advertising interaction or the outcome itself, independently of time-zone boundaries.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.