Picture two conversion actions tracking the exact same sales. One reports 5 conversions. The other reports 10. Which number do you trust?
Most people pick 10. Bigger has to be better, more complete, closer to the truth. Here is the uncomfortable part: in a lot of setups, the bigger number is the thinner one, and your bidding algorithm already knows it even if your dashboard does not.
What Consent Mode v2 Actually Does
When a visitor declines ad cookies and data sharing, you legally cannot tie that person’s purchase to a specific ad click at the user level. The identifiers that would connect click to sale are gone, by their choice.
Google’s answer is Consent Mode v2. Instead of going completely dark for that visitor, the browser still sends a stripped-down, identifier-free ping. Google takes those anonymous signals from everyone who declined and models the conversions in aggregate, estimating how many sales that opted-out group probably produced.
That word, models, is the whole story. Modeled conversions are not recovered sales. They are statistical estimates.
Decompose the Bigger Number
Go back to the 5 versus 10. The action showing 10 is not 10 matched sales. Break it apart and it is usually something like 5 observed conversions (real, but attributed by click and session only) plus 5 modeled conversions (an estimate of the opted-out cohort).
So 10 equals 5 real plus 5 educated guesses. The action showing 5 is 5 deterministic, user-matched sales you can point to. One number is a measurement. The other is a measurement stapled to a forecast.
Modeled Conversions Have Real Limits
Estimates have their place, but you have to know what you cannot do with them. You cannot point a modeled conversion at a specific order. You cannot dedupe it. You cannot reliably optimize at the keyword or audience level on top of it, because there is no individual event underneath, just an aggregate guess.
Modeled data is directional. It is fine for a rough sense of total volume. It is shaky ground for the granular decisions that actually move an account.
Why Smart Bidding Prefers the Smaller, Cleaner Number
Here is the counterintuitive payoff. Google’s own bidding algorithms weight observed, user-matched conversions far more heavily than modeled ones. Five clean, deterministic signals train the bidder better than ten noisy, partly-estimated ones.
Feed Smart Bidding a pile of modeled conversions and you add noise to the exact signal it optimizes against. A bigger input is not a better input. Quality beats volume, and the algorithm is built to know the difference.
You Cannot Merge the Two Cohorts (and Should Not Try)
The instinct, once you see the split, is to route the opted-out conversions back into the clean, identifier-based track so everything lands in one rich number. Do not. You cannot do it compliantly, and you cannot really do it technically either.
The clean track carries identifiers: a hashed email, a click ID. Pushing an opted-out person’s sale through it means sharing the identifiers they explicitly refused, which is the precise thing an opt-out prohibits. That is a privacy violation, not a clever workaround.
It is also a technical dead end. The identifier-based upload needs an identifier to attach the conversion to. For an opted-out user there is nothing to attach, so there is nothing to send. The only path that works for that cohort is the identifier-free ping that feeds modeling.
In other words, the split between a rich opt-in track and a thin opt-out track is not a gap to close. It is the compliance architecture doing its job. The good news: opted-out sales still reach Google, thinly, through modeling. They are not lost. They are counted the only way the law and the plumbing allow.
What to Actually Do
Judge conversion data by fitness for the job, not by size. Use deterministic, observed conversions for bidding and for any decision you are going to act on. Treat the modeled total as a directional ceiling: useful context, not a target.
And if you want to grow the number of conversions you can actually measure, the lever is your consent rate, the share of visitors who opt in, not a hack that smuggles opted-out data into the wrong pipe. Better consent UX expands the measurable pool honestly. Nothing else does.
A conversion count you cannot trace, dedupe, or bid on is not better just because it is bigger. Sometimes the smaller column is the honest one. For more on why GA4’s modeled layer carries weight it was never built for, see GA4 Was Supposed to Fix Everything. It Didn’t.