An attribution window defines how long after an eligible advertising interaction an outcome may receive credit under the reporting rules. Widening the window can change the reported result without changing what the ChatGPT campaign actually delivered. That makes window selection a measurement convention with business consequences.
The useful question is not “Which window gives the best ROAS?” It is how sensitive the campaign conclusion is to a clearly stated reporting choice. A disciplined comparison keeps everything else fixed and shows the effect of that choice rather than turning a favorable setting into a claim of improved performance.
Freeze the rest of the report
Record the account, campaign IDs, date interval, time zone, time basis, goal definition and event selection. Retrieve the compared views as close together as practical and keep their extraction timestamps. If recent data is still updating, the retrieval timing becomes another possible explanation for differences.
Change only the window being investigated. Do not widen the click window, add view-through credit and switch from conversion time to ad-event time in one step. That combined change produces a new result but does not reveal which setting caused it.
The dedicated conversion endpoint accepts click windows of 7, 14 or 30 days and view windows of 0 or 1 day. Omitting these parameters, or sending null, uses 30 days after a click and 1 day after a view. Set the view window to 0 for a click-only comparison. These reporting choices do not change campaign goals or optimization settings; OpenAI reporting documents the distinction.
Build a sensitivity table
Use hypothetical values to illustrate the structure. Suppose the same completed interaction period shows 18 attributed purchases under a shorter supported click window and 24 under a longer one, with view credit excluded in both. The difference is six purchases credited under the broader timing convention.
If spend is SEK 1,200 for the same campaign scope, cost per attributed purchase is about SEK 66.67 in the first view and SEK 50 in the second. The campaign did not become cheaper because you changed the report. The count receiving credit changed, and the ratio followed.
Show both results, the exact windows and the difference. If purchase values are available and appropriate to the question, compare value separately rather than assuming the six additional purchases have the same average value as the original eighteen.
Same campaign, different credit rule
- Shorter window
18 credited purchases give about SEK 66.67 per purchase.
- Longer window
24 credited purchases give SEK 50 per purchase.
- Interpretation
Spend did not change; more outcomes received credit.
Interpret the difference conservatively
A larger reported total under a longer window can be consistent with later outcomes, but the report alone does not establish why customers took longer or whether advertising caused those purchases. Avoid treating every additional attributed order as incremental revenue.
The sensitivity view can reveal dependence on the reporting convention. If the campaign meets an internal efficiency criterion only under the broadest view, that dependency belongs in the recommendation. It does not automatically mean the campaign is bad; it means the economic conclusion needs the convention to be explicit.
Conversely, similar results across supported windows do not prove attribution is causally correct. They show limited sensitivity to this particular parameter for this particular dataset. Keep the claim at the level the comparison actually tested.
Match the buying cycle without inventing certainty
A business with a longer consideration process may reasonably want to inspect later outcomes. Use observed timing evidence where available. A sales narrative that customers usually take weeks should be treated as a hypothesis until supported by a relevant dataset.
Do not select a window solely because it maximizes the reported result. Agree the primary convention before evaluating campaigns and retain alternative windows as sensitivity views. That consistency makes comparisons across reporting periods less vulnerable to cherry-picking.
If the business changes its primary convention, document the effective date and decide how to handle prior periods. Recalculating comparable history can be useful when supported. Mixing old and new conventions in one trend without a note is not a valid performance comparison.
Keep recent interactions from confusing the test
The newest clicks may not yet have had enough elapsed time to benefit from a longer window. A short-versus-long comparison on very recent activity can therefore look identical simply because later outcomes have not happened yet.
Choose an interaction period with the observation maturity needed for the question, or label the comparison provisional. This is distinct from the configured window length. A report can permit later attribution while the selected interactions remain too recent to have accumulated those outcomes.
Time-basis selection explains how dates organize the result, while click and view components keeps interaction categories separate. For causal limits, read attributed versus incremental revenue. Use OpenAI reporting for supported settings. The final decision should state the chosen convention, the observed sensitivity and the unresolved assumptions, not simply quote the most favorable number.
Sources and scope
Run a controlled sensitivity comparison of reporting windows while keeping campaign goals, time basis and source period fixed.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.