Modeled conversions, reported as if they were measured
A case study reports a conversions total. Say it is 1,400. The instinct is to read that as a count of things that happened. On Google’s ad platform, which documents the practice openly, that instinct is wrong often enough to be worth checking.
What the “conversions” number on an ads dashboard actually contains
Google’s documentation is direct. Its page on modeled online conversions describes them as estimates of conversions Google is unable to observe directly, then states plainly: “In the ‘Conversions’ column, Google reports both modeled and observed conversions.” The consent mode page adds that they “will appear in the ‘Conversions’ column and be reflected in all downstream reports that use this data”. Every metric on that column inherits the blend: cost per conversion divides spend by a partly estimated denominator, and ROAS rests on a conversion value carrying modeled values.
This is not measured versus wrong. Modeling is disclosed and deliberate, and without it reported conversions would show only the observable slice. The gap is in the case study, which passes the number along as a tally.
Measured and modeled, side by side
| Measured | Modeled | |
|---|---|---|
| What it is | An event the pixel, tag or API directly recorded, tied to an observed ad interaction. | A statistical estimate built from patterns in other, observable journeys. |
| What could make it wrong | Implementation errors, a tag firing twice on a reload, duplicate transaction IDs. | Thin training data, or its assumption that unobserved users behave like observed ones not holding here. |
| Where it shows up | The same field. Google states modeled conversions are reported at the same granularity as observed ones, and in the same “Conversions” column. No default visual separation marks which is which. | |
Why the platform is filling in a gap at all
The hole exists before the modeling does. Two documented causes:
- Declined consent. Refuse ads or analytics cookies and the link between ad interaction and outcome is gone. Google Analytics says behavioral modeling for consent mode “uses machine learning to model the behavior of users who decline analytics cookies based on the behavior of similar users who accept analytics cookies”, trained on the property’s own observed users.
- Browser and app restrictions. Google’s modeled conversions page names Safari and Firefox as browsers that do not allow conversion measurement using third-party cookies. Traffic affected by Apple’s App Tracking Transparency is modeled too.
The question is not whether a modeled component exists, but how big it is.
A hypothetical case, worked through
The numbers below are hypothetical, invented to show the arithmetic. They match no case study in this library and no real agency or brand. Picture an excerpt reading: “Across a 90 day flight on a $17,500 budget, the campaign delivered 1,400 conversions at a cost per conversion of $12.50.”
Now suppose the hypothetical split is 900 measured, 500 modeled. The headline does not change; what you can do with it does. The measured 900 is a floor you can reason about, and against it the observed cost per conversion is nearer $19.44. The other 500 is the platform’s estimate, and it may be a good one. But planning your own budget on $12.50 imports someone else’s consent rates and browser mix as though they were measurement.
The question that gets you the split
The question is short and not hostile: what percentage or count of these reported conversions is modeled rather than directly observed, for this specific reporting period? That last clause matters: consent rates and browser mix drift, so a figure from another quarter is no answer for this one.
A satisfying answer is specific: a figure, or a screenshot of the platform’s own reporting where that distinction is exposed. A reply like “about a third, mostly EU consent and Safari traffic” is a good one, because it names both figure and mechanism. A non-answer is reassurance that never touches the modeled share: the tracking was implemented correctly, the numbers came straight from the platform. Both can be true of a half estimated total. This is what actually backs up a case study number, narrowed to one thing.
When the agency does not know, or will not say
Two situations produce the same silence. The first is that the agency does not track the distinction, which is common and not a red flag on its own: the conversions column is what bidding optimises towards, and many teams have never decomposed it. The second is an agency that knows and answers around it, where the tell is redirection rather than refusal.
Either way, treat the number as directionally useful, not as a precise, fully observed count. Modeling is only one undisclosed choice behind such a figure, since the attribution window is also a choice, and rarely disclosed. For the wider checklist, see how to read a marketing case study without being misled.
See how the numbers are actually reported
Every case study here is republished with full credit to the agency behind it, and every figure belongs to that agency. Browse the case study library, and if you have work you would let someone read this closely, submit your own.
FAQ
Is a modeled conversion the same thing as a fake conversion?
No. It is a statistical estimate built from observable patterns elsewhere in the same account, not a fabrication, and Google documents the practice openly. The error worth catching is treating an estimate as identical in kind to a recorded event.
Can I see the modeled share myself, without asking the agency?
Not from a case study or a screenshot, and generally not without account access. Google Analytics documentation describes a data-quality icon and a Blended reporting identity setting that indicate whether modeled data appears in reports, but that is the Analytics property, not the ads conversions column. Neither Google Ads page cited below describes a self-serve report or segment that separates modeled from observed conversions; both say only that the two share one column. Asking remains the only route.
Does this apply to a single platform or all of them?
The evidence here is Google’s own documentation about Google’s ad platform, the only platform this piece can speak to. How widely the same thing happens elsewhere is not something to assert, since no verified count exists. Put the question to whichever platform the case study drew its number from.
Sources
- Google Ads Help, “About modeled online conversions”, fetched 6 September 2026.
- Google Ads Help, “About consent mode modeling”, fetched 6 September 2026.
- Google Analytics Help, “[GA4] Behavioral modeling for consent mode”, fetched 6 September 2026.
