Blended, Observed, or Device-Based: The GA4 Setting That Rewrites Your Numbers
GA4’s reporting identity setting decides whether your reports include modeled data, and switching it silently rewrites your user and conversion numbers. The default, Blended, fills gaps for non-consented users with Google’s modeling, so it usually shows higher counts. Observed and Device-based drop the modeling and report only users actually seen, so they show lower but more literal numbers. Neither is wrong, but they answer different questions. The way to trust the Observed view is to widen what GA4 can observe, by sending first-party events server-side so real users arrive as data, not estimates.
What the reporting identity setting actually does
It decides how GA4 stitches a user together, and whether it’s allowed to guess.
Buried in GA4’s admin is a setting most people never touch: reporting identity. It controls how GA4 decides that two visits belong to the same person, and crucially, whether it’s allowed to fill in the gaps with estimates. The default is Blended, and Blended injects Google’s modeled data to account for users it couldn’t directly observe. Observed and Device-based, by contrast, report only the users GA4 actually saw.
That single choice changes the numbers on your dashboard. Per OptimizeSmart, Blended tends to show higher counts because of the modeling, while Observed and Device-based show lower, more literal figures. Same traffic, different story, depending on a dropdown.
GA4’s default Blended reporting identity injects Google’s modeled data to fill gaps for non-consented users, so it typically reports higher user and conversion counts than Observed or Device-based.
Why your numbers move when you switch it
You’re not changing the traffic, you’re changing how much of it is estimated.
Here’s the thing: switching reporting identity doesn’t change a single real visit. It changes how much modeling gets mixed in. Google Analytics Help describes Blended as resolving identity in order, User-ID, then Google Signals, then Device ID, then modeling to cover the rest, so the modeled layer is exactly what disappears when you move to Observed.
Modeling also isn’t guaranteed. Google’s consent-mode modeling needs roughly 700 ad clicks per week per country before it produces any modeled data at all. Below that threshold, smaller sites may get little or no modeling benefit, which means Blended and Observed can look nearly identical for them, and very different for a large advertiser.
You may be interested in: The 70% problem: your AI traffic is hiding in GA4 ‘direct’
Blended vs Observed vs Device-based, side by side
Three settings, three different answers to the same question.
The practical differences come down to what each option counts and whether it estimates.
| Property | Blended (default) | Observed | Device-based |
|---|---|---|---|
| Uses modeled data | Yes | No | No |
| Uses User-ID and Google Signals | Yes | Yes | No |
| Counts only users actually seen | No | Yes | Yes |
| Tends to report | Higher | Lower | Lowest |
With EU consent rejection commonly reported between 40 and 70 percent, a large share of sessions go unobserved, which is precisely the gap Blended tries to model and Observed simply leaves out. The more consent you lose, the wider the spread between these three columns becomes.
Switching GA4 reporting identity doesn’t change reality, it changes how much modeling is mixed in, which is why your numbers move the moment you flip the setting.
Which number should you actually trust
The question isn’t which is correct. It’s which one you can defend.
Neither view is wrong, but they answer different questions. Blended estimates total reality and is useful for trend-watching; Observed reports literal, defensible data and is better for reconciliation and auditing. The catch is that as the unobserved share grows, Blended leans harder on estimates you can’t verify.
And that share is about to grow. Search Engine Land reports that from June 15, 2026, ad_storage is the sole control over GA4-to-Google-Ads data, which raises the proportion of users modeling has to cover. The more your reporting depends on modeling, the more a settings dropdown quietly shapes the numbers your decisions rest on.
Google’s consent-mode modeling needs roughly 700 ad clicks per week per country before it produces any modeled data, so smaller sites may see little benefit from Blended.
How to make the Observed view trustworthy
Stop arguing about modeling and widen what GA4 can actually see.
If Observed feels too low, the real fix isn’t to switch back to Blended and accept the estimates. It’s to let GA4 observe more of your real users in the first place. When events arrive as genuine first-party observations, the Observed view moves closer to reality and your reliance on modeling shrinks.
Transmute Engine™ sends first-party events server-side, so users who’d otherwise be lost to consent gating or browser limits arrive in GA4 as observed data rather than gaps. The inPIPE layer captures those events on WordPress, which is what lets the Observed identity report real numbers instead of estimates.
You may be interested in: Self-hosted vs managed: which WooCommerce event pipeline fits your store
Key Takeaways
The short version for your next analytics review.
- Blended is the default and it models: it fills non-consented gaps with estimates, so it reports higher.
- Observed and Device-based don’t model: they count only users actually seen, so they report lower.
- Switching the setting moves the numbers: it changes how much modeling is mixed in, not the traffic.
- Modeling has a threshold: roughly 700 ad clicks per week per country, or no modeled data appears.
- Server-side first-party data is the real fix: it makes the Observed view trustworthy by widening what GA4 can see.
Blended uses User-ID, Google Signals, and Device ID, then fills the rest with Google’s modeling. Observed uses the same signals but no modeling, so it counts only users actually seen. Device-based ignores User-ID and Signals and counts by Device ID alone.
Because the setting changes how much modeled data is mixed into your reports, not the underlying traffic. Blended adds modeled users to cover non-consented sessions, so moving to Observed or Device-based usually drops your counts to the users GA4 actually observed.
Compare the same report under Blended and Observed identity over the same date range. The gap between them is, broadly, the share that modeling is contributing. A large gap means you’re leaning heavily on estimates rather than observed data.
Server-side first-party events let GA4 actually observe users who would otherwise be lost to consent gating or browser restrictions. When real users arrive as observed data, the Observed identity reports closer to reality and you depend less on modeling.
References
- Google. “Reporting identity,” Google Analytics Help, 2026. https://support.google.com/analytics
- OptimizeSmart. “The best reporting identity for Google Analytics 4,” 2026. https://optimizesmart.com/blog/the-best-reporting-identity-for-google-analytics-4/
- Search Engine Land. “Google simplifies analytics and ads consent rules,” 2026. https://searchengineland.com/google-simplifies-analytics-and-ads-consent-rules-474206
- Google. “About consent mode behavioral modeling,” Google Analytics Help, 2026. https://support.google.com/analytics
If your GA4 numbers and your actual results don’t line up, see how Seresa makes the observed data real.