GA4 Reporting Identity: Blended vs Observed vs Device-Only Explained

Quick answer: GA4 lets you pick one of three reporting identity modes — Blended, Observed, or Device-only. They differ in how aggressively GA4 stitches user activity together across devices and sessions. Blended uses User-ID, Google signals, device, and modeling in that order. Observed drops the modeling step. Device-only ignores everything except the device-specific ID. The default is Blended, which gives the lowest user counts but the most realistic ones — provided you trust modeling.

This setting is one of the most quietly impactful in GA4: switching modes can change your user count by 10–30% on the same property over the same time period. Below is what each mode actually does, when to use which, and what changes downstream when you switch.

Why GA4 needs three modes

A "unique user" sounds simple — count the distinct humans. In practice, GA4 has access to several different identifiers, each with different reliability and different privacy implications. The reporting identity setting tells GA4 which identifiers to use, and in what order, when deciding whether two sessions belong to the same user.

The four identifiers in order of priority:

1. User-ID — an identifier you send on logged-in events. The most reliable, because you set it.
2. Google signals — Google's cross-device data, available for users who opted into ads personalization on their Google account.
3. Device — the client_id cookie, scoped to a single browser on a single device.
4. Modeling — when none of the above is available, GA4 uses machine learning to model what the user count probably is.

Blended

Uses all four: User-ID → Google signals → Device → Modeling, in that order.

This is the default. It's the most aggressive at stitching together what GA4 believes is the same user across devices and sessions. Modeling fills in the gaps where consent or signals are missing.

Best for: properties where you want cross-device unification and accept that some of the data is modeled rather than observed. Most ecommerce and consumer-facing sites fit here.

Caveats:

• The modeled portion can be a meaningful share of your user count — often 5–20% — and isn't separable in reports.
• Modeling only kicks in when consent is properly signaled. If your Consent Mode v2 implementation is broken, modeling produces less useful estimates.
• Engagement and revenue metrics aren't modeled; only user counts are.

Observed

Uses User-ID, Google signals, and Device. Drops modeling.

Observed gives you only data GA4 actually saw. No machine-learning estimates fill in gaps. If a user denied consent and their session wasn't fully recorded, they don't show up.

Best for: teams that need to defend their numbers to a CFO or auditor and don't want to explain modeled estimates. Also a reasonable default for regulated industries where modeled data is harder to justify.

Caveats:

• Your user counts will be lower than Blended — sometimes substantially.
• Cross-device unification still happens via User-ID and Google signals, just without the modeling layer.
• You're still subject to GA4 thresholding on small audiences regardless of identity mode.

Device-only

Uses only the device-level client_id cookie.

The most conservative mode. A single human using a laptop, a phone, and a tablet shows up as three users — because GA4 never tries to stitch them together. No User-ID, no Google signals, no modeling.

Best for: properties operating under strict privacy regimes, or organizations that have made a deliberate decision to treat cross-device identity as out of scope. Very common in EU public-sector and healthcare implementations.

Caveats:

• User counts will be the highest of the three modes — but they'll represent device sessions, not actual humans.
• Cross-device conversions look weak. A user who browses on mobile and converts on desktop appears as two separate journeys.
• Most attribution analysis becomes less useful, because the conversion attribution can't see the upstream cross-device touches.

Side-by-side comparison

How to pick

A practical heuristic:

• You're a typical SaaS, ecommerce, or media site with normal consent rates and decent User-ID coverage on logged-in flows → Blended. The modeling fills in the gaps and your reports are most useful.
• You're reporting numbers to a CFO, board, or external auditor, or you're nervous about defending modeled estimates → Observed. Lower counts but every number is defensible as something GA4 actually saw.
• You operate under strict privacy rules (public sector, healthcare, regulated education) where cross-device identification is out of scope → Device-only. Accept that "users" means "devices."

The biggest mistake here is picking a mode without thinking about it. The default is Blended, which is fine for most teams, but it's worth a deliberate decision because the choice affects every report downstream.

How to change it (and what changes downstream)

1. Open GA4 → Admin (gear icon).
2. Under Property, click Reporting Identity.
3. Pick the mode you want.
4. Click Save.

What changes immediately:

• All user-count metrics in standard reports recompute on the new identity, including historical data. Reports that previously showed 100,000 users may show 85,000 (or 115,000) the next time you refresh.
• Audiences re-evaluate against the new identity. Audience sizes can shift.
• Attribution paths can change, because the unification of touchpoints across sessions depends on the identity mode.

What doesn't change:

• Raw event data in BigQuery is untouched. You can recompute users any way you want from BigQuery if you want a reference point. See our data retention guide for why BigQuery export is worth enabling.
• Conversion counts (the events themselves) don't change. Only the user counts attached to them.
• Engagement and revenue metrics are session/event-level, not user-level, so they're stable across modes.

A note about retrospective comparisons

If you switch modes mid-quarter, you'll get a confusing discontinuity in user counts. Either switch at a clean quarterly boundary, or annotate the change in your reporting docs so the team knows why the line jumped. The change applies to historical data too — there's no flag saying "this was reported under mode X."

Frequently asked questions

Which reporting identity is the default?

Blended. If you've never changed the setting, your property is using it.

Can I see how much of my user count is modeled in Blended?

No — GA4 doesn't expose the modeled fraction in standard reports. The only way to estimate it is to temporarily switch to Observed and compare user counts for the same date range. Switch back when done.

Does reporting identity affect Google Ads conversion data?

Indirectly. The conversion event count itself is identity-agnostic, but the user-level attribution paths that determine which channels get credit can shift when you change modes. If you've imported GA4 conversions to Google Ads for bidding, expect some movement when you switch.

Is changing reporting identity reversible?

Yes. You can switch back at any time and your reports will recompute. There's no destructive change to underlying data.

Does the choice affect data thresholding?

Only marginally. Thresholding kicks in on small audiences regardless of identity mode — though Blended's modeling can occasionally push a borderline audience above the threshold. See our thresholding deep dive.

Find identity-mode mismatches automatically

Reporting identity is one of the property-level settings the Snifflytics audit checks — along with data retention, time zone, currency, and internal traffic filters. Run a free audit to see what's set, whether it matches your business stance, and what to consider before changing it.