Add up what Meta, Google, and TikTok each claim they converted last month. Now open your store's order count. If the platform sum exceeds your actual orders, congratulations: your measurement is working normally. That is the system functioning as designed, and understanding the design is the difference between reading ad reports and being played by them.
Four mechanisms produce the inflation. None is fraud. All of them happen to point the same direction, which tells you something about who designed them.
Mechanism 1: attribution windows built for generosity
A platform counts a conversion if the buyer interacted with an ad within its attribution window: commonly configurations like seven-day click, and historically longer. Your customer clicked a Tuesday ad, thought about it, searched you on Friday, and bought. The platform books the conversion; so does whatever channel closed on Friday. Both entries are "true" under their own rules. Stack a few channels with overlapping windows on one considered purchase and a single order becomes several claimed conversions before anyone has lied.
Windows are configurable, and shortening them makes reports humbler and optimization data thinner; the platforms default to the settings that flatter the platform, which is your first hint about how to read everything else.
Mechanism 2: view-through conversions
Meta and others can count a conversion when the buyer merely saw an ad (no click) within a shorter window before purchasing. On paper this credits genuine awareness effects, which exist. In practice it lets a platform bill for customers who scrolled past an ad on the way to buying for reasons the platform never touched: your newsletter, a friend's recommendation, an existing habit. Retargeting campaigns feast on this: they show ads to people already close to buying, then claim the purchase as ad-driven. View-through is the single most quietly inflationary setting in the stack, and it is on by default in more reports than people realize.
Mechanism 3: modeled conversions
Since iOS privacy changes and consent-driven data loss, platforms fill their observation gaps with modeled estimates: statistically inferred conversions from users they could not directly track. The models may well be roughly right on average; the point is that a growing share of "conversions" in your dashboards are estimates produced by the entity being paid for the results, with methodology you cannot inspect. When the party that profits from the number also models the number, the honest posture is calibrated skepticism, not outrage or faith.
Mechanism 4: everyone claims the whole order
Attribution inside each platform is solipsistic: Meta does not know Google exists in your funnel, and vice versa. Each platform that touched the buyer claims full credit under its own window and rules. Your GA4 property tells a third story, with its own model, and none of the three is summable with the others. The order count in your store is the only number in this paragraph that corresponds to money.
How to read platform numbers without being played
- Never sum across platforms. The platform totals answer "what did this platform touch," not "what did advertising produce." The store answers what happened.
- Compute each platform's ratio to reality. Claimed conversions divided by actual orders in the same window, tracked monthly. The absolute inflation matters less than its stability: a steady ratio means the instrument is consistent and usable for trends; a jumping ratio means something changed (settings, modeling, tracking) and needs investigation before decisions.
- Compare campaigns within one platform, not across platforms. Inside one platform's consistent rules, relative performance is meaningful. Across platforms, you are comparing different fictions.
- Audit the settings that inflate: attribution windows and view-through inclusion, per campaign type. Retargeting judged on view-through is a machine for buying your existing customers back with extra steps.
- For real budget decisions at real spend, incrementality beats attribution: pause a channel in a region and watch what actually changes. It is the only test the platforms cannot grade for themselves.
The platforms are not lying to you; they are answering a friendlier question than the one you asked, and answering it in their own favor. Your job is to keep asking your question, which is what the business actually earned. The gap between the two questions is measurable, stable, and honestly manageable, and if nobody in your company has measured it yet, that single afternoon of arithmetic (platform claims versus store orders, one month, one spreadsheet) will change more budget conversations than any new dashboard this year.