Meta Ads attribution inflation is bimodal: median 1.01x, but every purchase account sat at 2.2x to 4.4x
We asked one question of each of the 23 accounts: how many conversions does Meta report under the account's own attribution setting, against how many it reports under the strictest window it offers, 1-day click? Same account, same event, same lifetime window. Nineteen accounts returned a finite ratio. The median was 1.01x. Not one account was above 10x.
That median hides the shape. Fifteen accounts whose main event is a lead sat between 1.00x and 1.10x, with one exception at 1.39x. Four accounts whose main event is a purchase sat at 2.15x, 2.23x, 3.33x and 4.35x. There is nothing in between. Attribution inflation in Meta is not a percentage you can apply across a portfolio. It is a property of the event type.
The data is $783,518 of pooled spend (AUD) across 23 Australian accounts, pulled from the Marketing API v23.0 on 8 and 9 September 2026 with the action_attribution_windows parameter set to return both the default and the 1-day click figure side by side.
The 19 accounts
The default column is whatever the account's attribution setting produces, in most cases 7-day click plus 1-day view. The 1-day click column counts only conversions within 24 hours of a click on the ad. Inflation is the first divided by the second.
Acct | Band | Event | Default window | 1-day click | Inflation | At least 30 conversions |
|---|---|---|---|---|---|---|
S | Consumer goods & food |
| 283 | 65 | 4.35x | yes |
T | Consumer goods & food |
| 10 | 3 | 3.33x | no |
F | Manufacturing & equipment |
| 230 | 103 | 2.23x | yes |
A | Consumer goods & food |
| 1,752 | 815 | 2.15x | yes |
G | Home improvement & trades |
| 4,990 | 3,585 | 1.39x | yes |
V | Home improvement & trades |
| 32 | 29 | 1.10x | no |
P | Home improvement & trades |
| 2,135 | 2,024 | 1.05x | yes |
K | Professional & B2B services |
| 29 | 28 | 1.04x | no |
C | Manufacturing & equipment |
| 1,304 | 1,286 | 1.01x | yes |
E | Home improvement & trades |
| 178 | 176 | 1.01x | yes |
H | Health & care services |
| 254 | 252 | 1.01x | yes |
B | Manufacturing & equipment |
| 123 | 123 | 1.00x | yes |
D | Professional & B2B services |
| 400 | 400 | 1.00x | yes |
J | Home improvement & trades |
| 179 | 179 | 1.00x | yes |
L | Consumer goods & food |
| 268 | 268 | 1.00x | yes |
M | Home improvement & trades |
| 3 | 3 | 1.00x | no |
N | Professional & B2B services |
| 456 | 456 | 1.00x | yes |
Q | Health & care services |
| 32 | 32 | 1.00x | yes |
W | Health & care services |
| 193 | 193 | 1.00x | yes |
The four accounts not in the table (I, O, R, U) returned no lead or purchase event under either window, so there is no ratio to compute.
Read the Event column against the Inflation column and the split is total. Every purchase account is above 2x. Every lead account except G is at or below 1.10x. Account M reports omni_purchase at 1.00x, but on 3 conversions, which is below any floor we would use.
Account T, the 3.33x case, rests on 10 purchases under the default window and 3 under 1-day click. The ratio is in the right band for a purchase account, but 3 strict-window conversions is not a number to build on. We report it and set it aside.
The distribution
Measure | Value |
|---|---|
Accounts with a finite ratio | 19 |
Median inflation | 1.01x |
Above 1.3x | 5 |
Above 2x | 4 |
Above 10x | 0 |
Median with a floor of 30 strict-window conversions | 1.01x (n=15, min 1.00x, max 4.35x) |
Pooled: sum of default divided by sum of 1-day click | 12,851 divided by 10,020 = 1.28x |
The pooled ratio, 1.28x, is the number a portfolio report would show, and it is misleading in both directions. It overstates inflation for the thirteen lead accounts at 1.05x or below, and it understates it for every purchase account. Most of the pooled gap between 12,851 and 10,020 comes from two accounts: G, where the default window adds 1,405 leads, and A, where it adds 937 purchases.
A note on method, because it matters for the rest of the series. An early pass at this analysis produced ratios far above 10x. They turned out to be an artefact of comparing one action type under the default window against a different action type under 1-day click, the aliasing problem part 2 describes. Once each account was compared against itself, one event, two windows, the top of the range was 4.35x. The dramatic version was wrong, and we mention it because anyone reproducing this with a naive sum of the actions array will get the dramatic version.
Why this happens
Meta's attribution setting decides which conversions an ad gets credit for. The default for most accounts is 7-day click and 1-day view, which means a conversion counts if it happens within seven days of a click, or within one day of an impression with no click at all. The 1-day click window counts only the first of those, and only for one day.
Window | Counts a conversion when | Adds relative to 1-day click |
|---|---|---|
1-day click | Within 24 hours of a click on the ad | Baseline |
7-day click | Within 7 days of a click on the ad | Click-then-think conversions, days 2 to 7 |
1-day view | Within 24 hours of an impression, no click | View-through: conversions with no click at all |
7-day click + 1-day view (default) | Any of the above | Both of the above |
Whether those additions are large or zero depends on how the conversion can physically happen.
A pixel purchase happens on a website. A person can see an ad on Tuesday, not click it, search for the brand on Wednesday morning and buy. That is a 1-day view conversion. Or they can click on Tuesday, leave, come back through a bookmark on Saturday and buy. That is a 7-day click conversion. Both are invisible to 1-day click and both are ordinary shopping behaviour. Account S records 283 purchases under the default window and 65 under 1-day click, which means 218 of its purchases happened either more than a day after a click or with no click at all.
A Meta instant form lead happens inside Facebook or Instagram. The person taps the ad, the form opens, they submit. The lead is attached to that tap by construction. There is no path by which an impression with no click produces an instant form lead, and there is no path by which the lead arrives six days after the click, because the form is the click. So default equals 1-day click, and inflation is 1.00x. That is the story for B, D, J, L, N, Q and W, where the two columns are identical to the unit, and for C, E and H, where they differ by a handful.
A pixel lead is a website form, and it behaves like a purchase. Account G is the case: 4,990 leads under the default window, 3,585 under 1-day click. As part 1 showed, G's instant form composite is only 631, so most of its leads come through the website, and view-through and late-click conversions have somewhere to land. Account P, at 1.05x, sits between the two: 2,135 leads against an instant form composite of 1,711, so a minority of its leads are website forms and the inflation is proportionally small.
What it does to decisions
The attribution setting is not only a reporting choice. In Meta it is set on the ad set, and the delivery system optimises toward conversions as counted under that setting.
Decision | Under the default window | Under 1-day click |
|---|---|---|
Cost per purchase shown in Ads Manager, account S | Spend divided by 283 | Spend divided by 65 |
What the algorithm learns from, purchase accounts | Includes view-through and late clicks | Clicks that convert within a day |
What the algorithm learns from, instant form accounts | Identical either way | Identical either way |
Comparability with Google Ads or GA4 | Poor for purchases, fine for instant forms | Closer for purchases |
For a purchase account, cost per purchase under the default window is somewhere between a half and a quarter of cost per purchase under 1-day click. Neither is the truth. The default window credits Meta with purchases that a search click or a direct visit also touched; the strict window denies Meta credit for purchases that a Meta impression genuinely started. The point is not that one is right. It is that the two are 2x to 4x apart, and a monthly report that quotes cost per purchase without saying which window it used is quoting a number that could halve or double on a settings change.
For a lead account running instant forms, none of this applies. The two columns agree. Cost per lead is cost per lead. That is worth stating plainly, because lead generation advertisers often assume their numbers are the soft ones. In this set they are the hardest numbers Meta reports, and part 4 works through why.
The Google Ads sister series has the parallel: Google's lead form extension attributes on the click, and the same logic makes those leads resistant to window inflation.
Where it varies
By event type, not by industry. The bands only look different because they contain different event types.
Event type | Accounts | Inflation range | Median-ish behaviour |
|---|---|---|---|
| S, F, A (T below floor) | 2.15x to 4.35x | All above 2x |
| G | 1.39x | Behaves like a purchase, diluted |
| P | 1.05x | Small inflation |
| B, C, D, E, H, J, L, N, Q, W (K, V below floor) | 1.00x to 1.01x | None |
The three volume-floor accounts (T on 3, K on 28, V on 29 strict-window conversions) all land in the band their event type predicts: T with the purchases, K and V with the instant form leads. They do not change the picture; they just cannot be cited on their own.
What to check in your account
In Ads Manager, open Columns, then Compare attribution settings. Tick 1-day click alongside your default. Every conversion column now shows both.
Read the two figures for your main conversion event over the last 90 days and divide. If your event is an instant form lead, expect 1.00x. If it is a pixel purchase or a website form, expect well above 1x.
In the ad set, open Attribution setting. Note what it is. This is the number the delivery system is optimising to, and the number the Results column shows.
Through the API, request
action_attribution_windows=['default','1d_click']on the same insights call. Theactionsarray returns both values per action type, so you can compute the ratio for every event at once.In any report that leaves Ads Manager, add the attribution window next to every conversion figure. A cost per purchase without a window is not a number.
If your account mixes instant forms and website forms, split the lead count by
onsite_conversion.lead_groupedandleadbefore computing the ratio. The blended figure will sit somewhere in the middle and describe neither.
We run step 1 on every account in our Facebook Ads management reviews before quoting a cost per result, because the same account can be a 1.00x account for its leads and a 2x account for its purchases at the same time.
The principle
Attribution inflation is not a property of Meta, or of an industry, or of an account. It is a property of the conversion event: where it physically happens and whether an impression without a click can lead to it. Events that happen on Meta's own surfaces cannot inflate, because the click is the conversion. Events that happen on a website inflate by exactly as much as late-click and view-through behaviour allows, which for purchases in this set was 2x to 4x. Find out which kind of event you are reporting before you decide how much to trust the number.