Consent Mode: What GA4 Measures and What It Models

Consent Mode is often described as the thing that “recovers” the data the cookie banner takes away. That description is half right, and the wrong half is expensive. Consent Mode is a way for your banner to tell Google's tags what the visitor chose. What happens to the visitors who said no depends on how it is implemented, on how much traffic you have, and on which report you are looking at.

Incapto's team described the result precisely: “Consent Mode left us with a structural blind spot: we knew there was traffic we were not seeing, but we had no way to size it.” When they ran GA4 with Consent Mode next to Sealmetrics on their Shopify store for 48 days, GA4 did not record 29% of visits or 45% of pageviews. This piece explains which part of that gap Consent Mode can estimate, and which part it cannot.

The four signals, and the two implementations

Consent Mode carries four consent types: ad_storage and analytics_storage for cookies and device identifiers used for advertising and analytics, and ad_user_data and ad_personalization for sending user data to Google for advertising and for personalised ads. The last two are what version 2 added. The consent banner, your consent management platform, sets them; Google's tags read them.

How the tags behave depends on the implementation:

What is measured

For visitors who accept analytics cookies, GA4 works as it always has: users, sessions, events, sources and conversions, observed. For visitors who refuse, a basic implementation observes nothing. An advanced implementation observes that a hit happened, with its timestamp, browser, referrer and consent state, but without a cookie there is no identifier to tie one hit to the next, so those pings cannot be assembled into users and sessions the way consented hits are.

There is also a timing effect that applies to both modes. A visitor who accepts on the second page has already left the landing page that carried the campaign parameters, which is one of the reasons consented traffic still ends up in (direct) / (none).

What is modelled, and only above a threshold

GA4's behavioural modelling uses machine learning to estimate the behaviour of visitors who declined analytics cookies from the behaviour of similar visitors who accepted. It requires an advanced implementation on every page and a property that collects at least 1,000 events a day with analytics storage denied for at least 7 days, and at least 1,000 daily users with it granted for at least 7 of the previous 28 days. Google notes that meeting those prerequisites does not guarantee eligibility.

• What it estimates: users, sessions and new users, blended into standard reports with observed data.
• What it does not touch: event counts such as page_view and session_start in standard explorations.
• Where it does not exist: the BigQuery export and other data exports, audiences, user explorer, segments with sequences, retention reports and predictive metrics.

Google Ads runs its own modelling. When a visitor does not consent, modelled conversions appear in the Conversions column and in the reports built on it, provided the account reaches 700 ad clicks over seven days per country and domain grouping. That is the number bidding runs on, and it is Google's estimate of Google's own contribution. How that plays out against measured revenue is covered in how to measure ROAS after cookie consent.

How to see how much of your GA4 is modelled

1. Check the implementation. In Tag Assistant, load a page with consent denied. If no Google Analytics request leaves the browser, you run basic mode and nothing is modelled from pings.
2. Switch the reporting identity. In GA4's admin, compare the same report under Blended and Observed. Blended adds modelling; Observed does not. Switching changes only the reports, not the data collected.
3. Read the difference by metric. Users and sessions move; page views in explorations do not. A large gap in users with no gap in page views is modelling at work.
4. Compare with the BigQuery export. The export contains observed data only. If your warehouse and the GA4 interface disagree on users, this is usually why.
5. Reconcile against orders. Neither Blended nor Observed tells you how many visits really happened. Your real orders are the only total that no analytics tool produces; the method is in one number for marketing and finance.

What modelled data is good for, and what it is not

Modelled users and sessions are a reasonable way to keep a trend line from collapsing when consent rates change, and modelled conversions help Google Ads bid. They are not a substitute for measurement in three situations that matter to a budget owner:

• Channel decisions. Google's documentation describes estimated users, sessions and new users. It does not describe giving each unconsented visit its own source, and the loss is not even across channels: at Incapto, Sealmetrics recorded 11% more direct traffic than GA4 but 37–52% more from paid campaigns and 133% more from organic social.
• Anything downstream of the export. BI dashboards, data science and finance models built on BigQuery see observed data only.
• Sites below the threshold. A property that never reaches the prerequisites gets no behavioural modelling at all, only the consented share.

The alternative: measure without needing consent

The other way to close the gap is to stop depending on analytics cookies. Sealmetrics stores nothing on the visitor's device and reads the source of every landing page from its UTMs or referrer, so each visit and each conversion is observed with its channel rather than estimated. Nothing is modelled, and the same numbers are in the dashboard, the API and the export. The architecture is explained in complete data.

It has limits of its own, and they are the reverse of Consent Mode's. Sealmetrics does not identify users, so there are no retention reports, cohorts or cross-session journeys to estimate or observe. Ad blockers can still stop the tracker unless it is served from a subdomain of your own domain. And whether analytics without cookies is exempt from consent in your market depends on your configuration and your national authority's criteria.