How much of what Meta claims can you actually verify?

Meta says it drove almost every sale I made. How much of that can I actually verify?

Most attribution arguments end in a stalemate: the ad platform reports one number, the store reports another, and there is no shared basis on which either can be checked. So here is a measurement with its method attached, taken from one store's production data, and a test you can run on your own account.

What was measured

One Mexican ecommerce store, 90 days, from 7 June to 4 September 2026. Meta's ad account for that store reported 7,647 purchases in the window. The store recorded 7,712 orders in the same window: every order, from every channel, from every customer, new or returning.

Meta claimed 99.2% of them.

Why that is not an accusation

It would be easy, and wrong, to call this over-counting. The store's total order count is a hard ceiling on what all platforms combined can truthfully claim, and 99.2% is under it. Nothing here is provably impossible.

It is also genuinely possible for a platform to be right about a sale the store cannot see. Someone who blocks the pixel, or who never clicks and buys a day after seeing an ad, is a real buyer the store's own analytics will never attribute. Meta has better identity resolution than a first-party pixel does, and that advantage is not fraud.

Why it is still worth publishing

Because of what 99.2% does to the arithmetic. If one platform accounts for essentially every sale the store made, there is no room left for anything else to have contributed anything: not organic, not email, not direct, not a second ad platform. The claim is not impossible, but it is close to unfalsifiable from the platform's own reports, and a number you cannot check is a poor basis for deciding where the next peso goes.

The useful question stops being whether the claim is true and becomes how much of it you can verify, and what the rest is resting on.

Google could not be included, and that matters

The same run pulled Google's reported conversions for the same store and window: 17,484, or 227% of the orders that existed. That looks like a far bigger finding and it is not a finding at all.

Google's conversions metric sums every conversion action an account has configured, not purchases: add-to-cart, newsletter signups, phone calls, whatever the advertiser set up. Comparing it to order counts is comparing two different things and calling the difference a discovery. It was left out for that reason, and it is worth knowing before reading any comparison someone else publishes: if a benchmark quotes Google conversions against orders, it is measuring configuration, not attribution.

Meta's figure is comparable because the number pulled is specifically the purchase action, not a broader sum.

The test that does produce proof

Underneath all of this is one asymmetry, and it is the whole reason the exercise is possible. A platform can legitimately resolve a buyer you never saw. But no platform can resolve an order that does not exist.

So: take a window. Total the conversions every connected platform claims for it, on the platform's own conversion-date basis. Compare that against the orders your store actually settled in the same window. If the claims exceed the orders, the excess is not a modelling difference or a window artefact. It is double-counting, and it is provable without knowing which platform is responsible.

Four things will break the test if you skip them, and each one produces a confident wrong answer. Use the purchase action, not a platform's total conversion count. Pair each ad account with the stores it actually advertises for, not with one brand out of several. Do not run it on a window whose last day is still open, because settlement lag alone will manufacture impossible sales. And turn off any new-customer filter, which narrows your side of the comparison while leaving the platform's side whole.

What this store's number means in practice

Under the ceiling and near the top of it. Meta's contribution is not disprovable here, and it is also not independently confirmable from Meta's reports alone. Which is the situation most advertisers are actually in, and the reason attribution anchored on orders exists: not to name a winner between the platform and the dashboard, but to say which part of the claim is verified, which part is plausible and unverifiable, and which part, when the ceiling breaks, is arithmetic that cannot be true.

This covers one store and one platform over one window. It will be re-run and extended as more merchants come in and as they agree to be included; the method above is what makes a later version comparable to this one rather than a fresh set of numbers.