Attribution windows decide how long a conversion can look backwards to find something to credit, and most advertisers are running several at once without having chosen any of them. The defaults differ by campaign objective, so one account routinely reports on three bases simultaneously and averages them together.
That is not a settings mistake somebody made. It is what happens when you accept the defaults, which nearly everybody does.
This is for you if you compare campaign performance inside one Meta account and assume the numbers are measured the same way.
What attribution windows are doing
They set the reach-back period for credit.
A seven day click window means a conversion happening within seven days of a click gets attributed to that click. A one day view window means somebody who saw the advert without clicking, then converted within a day, also gets credited.
Widen the window and reported conversions rise, because more outcomes fall inside it. Narrow it and they fall. No behaviour changed in either case.
So the window is not a measurement of your advertising. It is a rule about which outcomes count as yours, and it sits underneath every efficiency figure you quote.
The default nobody chose
Here is what is actually running, observed live in accounts in August 2026.
Leads and sales objectives default to seven day click, one day view or one day engagement. Awareness and engagement objectives default to a one day window, appearing as one day click or as seven day click or one day view depending on the campaign.
The engagement component is worth noting on its own. Meta added an engaged view category, so some credited conversions involve neither a click nor a conventional view. Anybody working from older knowledge will describe a model that no longer exists.
None of that is a criticism of the defaults. Different objectives genuinely warrant different treatment, and the choices are defensible.
The consequence is the problem. Because the default varies by objective, a single account ends up running multiple attribution windows without anybody making that decision.
What Meta does with the mixture
It blends them, and it tells you, in a place almost nobody reads.
When your reporting spans campaigns using different settings, Meta’s own footer says Multiple attribution settings. Then it reports an average across them anyway.
So your account level return on ad spend can be an average of numbers measured three different ways. That is not a rounding difference or a margin of error. It is an apples-to-oranges aggregation, presented as a single figure, invisible unless you go looking.
The fix for visibility is one column. Add the attribution setting column to your reporting and you can see, per campaign, which basis produced each number. It is verifiable by anybody in their own account in under a minute.
I would put that column in every Meta reporting view as standard, and almost nobody has it.
One account, four different returns
Here is a live example from a US flag and banner ecommerce account, all from the same period.
| How you cut it | Return on ad spend |
|---|---|
| Mixed attribution windows, as Meta blends them | 1.11 |
| All campaigns including awareness | 1.25 |
| Sales ad sets only, lifetime | 1.58 |
| Sales ad sets excluding one failed geographic test | 1.99 |
Same account, same client, same period. Four answers, and every one of them is defensible.
The 1.11 is what somebody gets by opening the account and reading the top-level figure. The 1.99 is what the campaigns actually built to drive sales achieved, once a known failed test is excluded.
Anybody could report any of these honestly. That is precisely why the reporting basis has to be stated alongside the number, and why a figure quoted without it tells you less than it appears to.
Why shorter attribution windows are not safer
There is a common instinct that a shorter window is the conservative choice. It is not conservative, it is differently wrong.
A short window undercounts conversions that genuinely took longer to arrive. For a considered purchase, or any business where people compare before buying, that systematically understates the campaigns doing the persuading.
You have not removed bias by shortening. You have swapped an optimistic bias for a pessimistic one, and applied it unevenly across campaign types depending on how quickly each one’s audience decides.
The same logic applies in reverse to long windows, which credit clicks from weeks earlier on an assumption that gets less plausible with distance.
Neither is safe. One is generous, one is stingy, and consistency is the only property actually available to you.
There is evidence on how far this class of assumption can drift from reality. Gordon, Zettelmeyer, Bhargava and Chapsky compared observational attribution against randomised experiments across 15 Facebook experiments and 500 million user-experiment observations, and found estimates off by a factor of three in half the studies tested.
Attribution windows are one input into exactly that kind of observational estimate. Tightening or loosening one moves your reported number without moving it towards truth, because the underlying method carries far larger uncertainty than the window setting does.
That is a reason to stop treating window choice as a precision exercise. It is a reporting convention to hold steady, not a dial that gets you closer to the real answer.
What happens when you change one
Changing an attribution window is not free, and the cost lands in comparability.
Your reported numbers move immediately. Your history does not become directly comparable with your future, so the ability to say whether performance improved is what you have spent.
That trade is sometimes worth making. It is rarely worth making mid-quarter, and it is never worth making without telling whoever reads the reports first.
Export your existing figures before you change anything, and keep them somewhere the platform cannot restate. Ten minutes of work preserves what you actually told people last month.
What to do instead of optimising the window
Three things, none of which involve picking a better window.
Add the attribution setting column and find out what you are currently running. Most people discover a mixture they did not know about, and that discovery is worth more than any setting change.
Then compare like with like. When you rank campaigns against each other, check they share a basis, and note it when they do not.
Then state the basis whenever you quote a number upward. “1.58 on sales campaigns, seven day click” is a defensible sentence. A bare figure invites somebody to compare it against something measured differently.
What to change this week
Three steps.
Add the attribution setting column to your Meta reporting view and look at what your campaigns are actually using. That takes a minute and it is the whole point of this article.
Then check whether any comparison you routinely make spans campaigns on different settings, because that comparison is not measuring what you think.
Then write your reporting basis into the top of whatever report leaves your desk, so nobody downstream compares it against a number built on a different rule.
What conversion numbers actually measure covers the counting underneath, attribution models covers how credit gets divided once a window admits it, conversion lag explains why the window interacts with how long people take, and GA4 key events covers the naming confusion in the neighbouring tool. Getting events and parameters onto the page correctly is web development work as much as advertising work. To find out what basis your own numbers are on, book a teardown.