GA4 behavioral modeling for consent mode

This guide is built from Analytics Help: Behavioral modeling for consent mode, Reporting identity, Consent mode reference, Introduction to user consent management and Verify and update consent settings. Quotes are Google's wording; the last section is our own practice.

What behavioral modeling does

Once a site shows a consent banner, some visitors decline analytics cookies and GA4 loses the thread on them. Without a persistent identifier, Google explains, "if Analytics collects 10 page view events, it can't observe and report whether that's 10 users or 1 user." Behavioral modeling for consent mode fills that gap. It "uses machine learning to model the behavior of users who decline analytics cookies based on the behavior of similar users who accept analytics cookies."

The model learns from your own traffic: "The training data used for modeling is based on the observed user data from the property where modeling is activated." Google's examples of what it helps with are user and session questions such as daily active users, new users from a campaign, and how many visitors come from Germany versus the UK.

Google lists three safeguards. Models are checked with holdback validation, where estimates are compared to "a portion of observed user data that was held back from model training". Modeling is "only included when there is high confidence of model quality". And the general algorithm is applied separately to each business.

What your consent setup must send

Modeling works from the data that still reaches GA4 when consent is declined. Under consent mode, when analytics_storage is denied the tag stops reading and writing first-party analytics cookies. Google's consent mode reference continues: "Cookieless pings will be sent to Google Analytics for future measurement. Google Analytics will use cookieless pings for modeling." In apps, events without device or user identifiers are sent and used the same way.

That only happens in advanced consent mode, where Google tags load before the banner and send cookieless pings while consent is denied. In basic mode nothing is sent before the visitor chooses, so there is nothing to model from. Our cookie banner guide explains the two modes and the legal side, and the GTM consent settings guide covers the container setup.

The framework matters too. Google says: "When your users deny consent with a consent solution that uses the TCF, Google Analytics properties can't model data to fill in the missing information." So a banner that only passes an IAB TCF string, without Google consent mode, gets no modeling.

Eligibility: the thresholds

Because the model trains on your property's observed data, the property needs enough of it. Google lists these prerequisites:

Requirement	What Google says
Coverage	Consent mode is enabled "across all pages of your sites and/or all app screens of your apps."
Implementation	Tags load before the consent dialog appears and "Google tags load in all cases, not only if the user consents (advanced implementation)."
Declined traffic	"The property collects at least 1,000 events per day with analytics_storage='denied' for at least 7 days."
Consented traffic	"The property has at least 1,000 daily users sending events with analytics_storage='granted' for at least 7 of the previous 28 days."

Meeting these does not guarantee it. Google says it may take more than 7 qualifying days to train the model, that even more data may not be enough, and that the model also weighs "the ratio of new to returning users and user to session counts."

There is no switch to request it. "Behavioral modeling is automatically enabled when a given property becomes eligible." If a property later drops below the prerequisites, estimated data stops; if it qualifies again, estimates return "only from the date the property became eligible again." The daily user figure is the one most small business sites will miss, so many properties never see modeled data at all.

Turning it on: Blended reporting identity

Eligibility makes modeling available. You only see it in reports when the property uses the Blended reporting identity, which Google defines as "By User-ID, device ID, then modeling." The steps:

1. In Admin, under Data display, click Reporting identity.
2. Select Blended.
3. Click Save.

You need the Editor role or above. Choosing Observed or Device based hides modeled data again, and Google says the option "does not affect data collection or processing" and can be switched "at any time without making any permanent impact on data." Our GA4 User-ID guide compares the three reporting identities and how they relate to data thresholds.

How to tell when modeled data is in a report

GA4 mixes modeled and observed data in the same numbers, and Google warns you will "probably see differences" against observed-only reports, "for example, higher user counts". The data-quality icon in a report tells you which you are looking at:

Icon status	Meaning
Including estimated user data	Since the modeling effective date, Analytics is "estimating data that's missing due to factors such as cookie consent." Estimates may not be ready for yesterday yet.
Excluding estimated user data	The reporting identity is not Blended, so reports "only include data available from users who consented to the use of identifiers."
Estimated user data unavailable	One of three reasons: the date range is before the property became eligible; the report uses retention data or a segment with a sequence; or the property doesn't meet the eligibility criteria.

The last row is the one to read carefully. An unavailable status because of the date range is normal right after eligibility starts; the eligibility message means the thresholds above are not met.

Reports vs explorations

Modeling is not applied the same way everywhere. In standard reports it is applied to "metrics like users, sessions, and new users counts" but "not applied to event counts such as page_view, first_visit, and session_start." Unconsented events become reportable, but because GA4 can't tell whether they come from the same person, a visitor who declined sends first_visit and session_start again on each page load. Those event counts can look inflated in reports.

Path and funnel explorations do the opposite: there, modeling is applied to first_visit and session_start, so "the first_visit and session_start event count is lower in path and funnel explorations than in reports." If the same event shows two different totals, this is often why.

In Explore, a banner tells you what is included. Google's wording: "Most templates include only data from users who consented to the use of identifiers, except for the free-form and segment-overlap templates, which do include data from estimated users." Our explorations guide lists the other reasons reports and explorations disagree.

Where modeled data never appears

Google lists features that don't use modeled behavioral data:

• Audiences
• The user explorer, cohort and user lifetime explorations
• Segments with a sequence
• Retention reports
• Predictive metrics
• Data export, for example the BigQuery export

The BigQuery point catches people out. The export holds observed events only, so user totals you count in BigQuery will not match a Blended report. Neither number is wrong; they answer different questions.

Checking the consent signals

GA4 shows what consent signals it receives per data stream. In Admin, under Data collection and modification, click Consent settings and select a data stream. The "Behavior analytics consent signals" section "shows the status of consent signals related to features that measure user behavior on your site", next to a section for the advertising signals. After you fix a banner or tag, give it time: "It may take 48-72 hours for the notifications to update." For the tag side, GTM preview mode shows the consent state on each event.

How we handle it

This part is our own practice, not Google guidance. On client sites with a consent banner we set up advanced consent mode when the client and their legal adviser are comfortable with it, because it keeps the option of modeling open. We then check the data-quality icon rather than assume modeling is on: most small business sites stay below the consented-traffic threshold, and then Blended behaves like Observed.

When a property does become eligible, user counts can step up from the modeling effective date. We add an annotation on that date so nobody reads the jump as a traffic gain, and we compare periods only when both sides include, or both exclude, estimated data. For anything built on audiences or BigQuery, we tell the client up front that those numbers are observed-only.

Frequently asked questions

Is GA4 behavioral modeling on by default?

It switches on automatically once a property meets Google's eligibility criteria, but you only see it in reports with the Blended reporting identity selected.

How much traffic do I need for behavioral modeling?

Google's prerequisites are at least 1,000 events a day with analytics_storage denied for at least 7 days, and at least 1,000 daily users with analytics_storage granted for at least 7 of the previous 28 days. Meeting them does not guarantee eligibility.

Does behavioral modeling work with basic consent mode?

No. Google's prerequisites require the advanced implementation, where tags load before the consent dialog and send cookieless pings when consent is denied.

Why are users higher in reports than in BigQuery?

With Blended identity and modeling active, reports include estimated users. Data export, including BigQuery, doesn't support modeled data, so it only holds observed events.

Can I turn modeled data off?

Yes. Choose the Observed or Device based reporting identity. Google says this doesn't change data collection or processing and can be switched back at any time.