What OpenAI's Attribution Windows Actually Count
By Ansh Khandelwal 8 min read
In brief
ChatGPT Ads attribution windows are configured per conversion event. Learn how that choice changes reporting and optimisation.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 October 2026
An attribution window in ChatGPT Ads answers one question: how long after an ad interaction will OpenAI still credit a conversion to it. Two windows exist. The click-through window is configurable. The view-through window is fixed at one day and cannot be changed.
That is the entire mechanism. What follows is what each one counts, where the window is actually set, and why the default is the wrong shape for a B2B cycle measured in weeks.
What counts as a click and what counts as a view
A click is a user selecting your ad and arriving at your destination URL with oppref appended. From that moment the click-through window runs. Any conversion the pixel or Conversions API reports inside it, joined back by the stored oppref value, is credited.
A view is an ad impression with no click. The view-through window is one day, fixed. OpenAI's documentation labels the metric VTA (1d), and the Ads Manager interface renders it as "View-through conversions (1 day)", which is the same number under two names and a small trap for anyone building automated reporting on column headers.
Where both apply, the click wins. OpenAI's documentation is explicit that a conversion qualifying for both treatments is credited to the click. So the two windows do not double-count, and your total conversions figure is not the sum of two overlapping populations.
The view-through window was added through documentation updates on 18 and 19 August 2026, without a blog post or press release, as reported by PPC Land. Treat the exact date as reported rather than confirmed.
Where the window is actually set
Not on the campaign. Not on the ad group. On the conversion event setting.
The event setting is the object a Conversions objective campaign optimises toward, and it takes four fields: name, event_type, attribution_window_days, and source_ids. The window lives in that third field, bound to one event and one source.
Three consequences follow, and each of them surprises someone.
First, two campaigns optimising toward the same event share one window, because they share one event setting. You cannot run a 7-day window on brand and a 45-day window on prospecting for the same conversion event.
Second, you can run different windows for different events. lead_created and subscription_created are separate event settings with separate attribution_window_days. This is the lever that makes the rest of this article workable.
Third, changing the window changes what the optimiser sees, not only what the report says. In a Conversions objective campaign, the conversions inside the window are the training signal. A window too short for your cycle does not merely under-report. It teaches bidding that the traffic does not convert.
The default is wrong for B2B, and here is the arithmetic
OpenAI does not publish a documented default window value or the permitted range. Interface screenshots circulating in 2026 show a 30-day click window, with no counterpart in published guidance, and OpenAI's own text refers only to "the applicable configured window for each account". So the honest position is that you should read your own account's setting rather than assume a figure from anyone's screenshot, including one in a well-sourced trade report.
Whatever the number is, the shape of the problem is the same. Take a B2B offer where the median time from first touch to demo booked is 19 days and to closed-won is 74 days.
| Window | What lands inside it | What the optimiser learns |
|---|---|---|
| 7 days | The fastest-moving fraction of demo bookings only | That most clicks are worthless, and that the small segment which converts within a week is the whole market |
| 30 days | Most demo bookings, almost no closed-won | That demo booking is the outcome, which is fine if demo booking is the event you set |
| 90 days | Demo bookings and most closed-won | Reporting that keeps restating itself for a quarter and a feedback loop too slow to optimise a 25 USD per day budget |
The error to avoid is picking one window for everything. The workable configuration is two event settings.
Set a short window, roughly matching your median time to the earliest reliable event, on the event you optimise toward. For most B2B that is lead_created or appointment_scheduled at somewhere between 14 and 30 days. That is the number the bidder learns from, and it needs to be fast enough to be a signal.
Set a long window on a later event you send from the CRM through the Conversions API and never optimise toward. That is the number you report ROI on. Its slowness does not matter, because nothing is bidding on it.
Why long windows are not free
There is a real cost to the long window, and the measurement advice in this category tends to skip it.
Numbers stop being final. A 90-day window means last month's campaign row keeps rising for a quarter. Any weekly reporting cadence built on it is reporting an incomplete figure and calling it a result. If you run a long window, you need a stated maturation point, for example "we do not judge a cohort before day 45", and you need to hold to it when a number looks bad on day 10.
Late credit is weaker credit. A conversion 80 days after a single ad click, on a channel with no user-level data and no adjacency reporting, is a join on a stored identifier, not evidence that the ad caused the purchase. A long window increases the reported number and decreases the confidence per unit of it. Both things at once.
It does not fix the dark funnel. The window governs conversions the pixel can see. The conversions that arrive through a later branded search or a direct visit on a different device are outside the pixel's reach at any window length. That is a separate problem and it is covered in our post on the conversions ChatGPT Ads will never show you.
One more operational point. OpenAI states attributed conversions may take 24 to 48 hours to appear in Ads Manager. That delay sits on top of your window, not inside it, so the real time from ad click to a number you can read is the window plus up to two days.
What we cannot tell you
- The documented default window value and the permitted range. OpenAI's published text refers to the configured window per account without naming either. The 30-day figure comes from interface screenshots, not documentation. Undocumented.
- Whether changing a window backfills historical conversions. Not documented. OpenAI does document that a mismatched event name will not backfill, which is a reason to assume nothing backfills until you have tested it.
- The first-party cookie lifetime for the stored
opprefvalue. Unpublished, so the true ceiling on pixel-only attribution may be shorter than any window you configure. - Whether view-through credit is deduplicated across multiple impressions. Undocumented.
- Any first-party window comparison from our own data. InPromptAds runs no campaigns and has no account in which to run one.
Quick answers
What is the ChatGPT Ads attribution window? The period after an ad interaction during which OpenAI still credits a conversion to it. The click-through window is configurable per conversion event setting. The view-through window is fixed at one day and cannot be changed.
Where do I change the attribution window?
On the conversion event setting, in its attribution_window_days field, not on the campaign or ad group. All campaigns optimising toward that event share the same window.
Does ChatGPT Ads have view-through conversions? Yes, at a fixed one-day window, added through documentation updates in August 2026. Where a conversion qualifies for both click and view credit, OpenAI credits the click.
What window should a B2B advertiser use? Two. A 14 to 30 day window on the event you optimise toward, so the bidder gets a usable signal, and a longer window on a CRM-sent later event that you report on but never optimise toward.
Why did my conversion count go up after I did nothing? Either you are inside the 24 to 48 hour reporting delay OpenAI documents, or conversions are still arriving inside an open attribution window. Long windows mean historical rows keep moving.
Does a longer window mean better measurement? It means a bigger number and weaker evidence per conversion. Credit assigned 80 days after one click on a channel with no user-level data is a join on a stored identifier, not proof of cause.
Sources
| Claim | Source | Tier |
|---|---|---|
| Configurable click-through attribution windows exist in Ads Manager | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
Event settings take name, event_type, attribution_window_days and source_ids, one source per setting |
OpenAI Developers, Conversion Setup API reference, 2026 | Confirmed, primary |
| View-through window fixed at one day; click takes precedence where both qualify; labelled VTA (1d) | OpenAI documentation, as reported by PPC Land, August 2026 | Confirmed, primary for the rule; Reported for the rollout date |
| No published default value or permitted range; interface screenshots show a 30-day click window with no documented counterpart | PPC Land, 2026 | Reported |
oppref is appended on click and stored in a first-party cookie |
OpenAI Help Center, Conversion Measurement, 2026 | Confirmed, primary |
| Attributed conversions may take 24 to 48 hours to appear; mismatched event names do not backfill | OpenAI Help Center, Measure Results, 2026 | Confirmed, primary |
| Two-event-setting configuration for B2B cycles | Inference, ours | Inference, ours |
Related reading
- How to Install the OpenAI Ads Pixel, Directly or Through GTM
- The OpenAI Conversions API, and When Server Side Beats the Pixel
- The Conversions ChatGPT Ads Will Never Show You
- Why ChatGPT Ads Traffic Lands in Referral in GA4, and How to Fix It
- How to Measure ChatGPT Ads, and What You Cannot Measure Yet
Changelog
12 September 2026, v1.0. First publication. Locates the attribution window on the conversion event setting rather than the campaign, and sets out the two-event configuration that separates the optimisation signal from the reporting signal.
Field kit
Tools
Site
OpenAI (primary)
Ansh works across GEO strategy, B2B research, and execution. At InPromptAds, he translates new AI advertising products into clear operating advice, tests, and measurement questions for marketing teams.
