How Google consent mode models conversions when users decline cookies

When a visitor declines cookies, advertisers are no longer able to directly tie that visitor's ad interactions to conversions. Consent mode is the mechanism Google built to fill part of that gap. It passes the visitor's consent choice to Google's tags, and Google's models then estimate attribution paths for the unconsented journeys.

Eligibility depends on how the tags are implemented, how much traffic the account has and how many visitors consent. The figures below come from Google's own documentation. Google gives one worked example of recovery and no general recovery rate.

Consent mode v2 added two parameters in November 2023, ad_user_data and ad_personalization, alongside the existing ad_storage and analytics_storage. The first covers sending user data related to advertising to Google. The second covers personalised advertising.

Basic and advanced consent mode behave differently when consent is denied

Google's developer documentation describes two ways to implement consent mode. In basic mode, Google tags stay blocked until the visitor interacts with the consent banner. If the visitor declines, nothing is sent to Google, not even the consent status.

In advanced mode, the tags load when the visitor opens the site or app, with consent defaulting to denied unless other defaults are set. If consent stays denied, the tags send the consent state and measurements without cookies. Full measurement data is sent only after consent is granted.

The modelling difference follows from that. Google states that in basic mode, conversion modelling in Google Ads relies on a general model, which is less detailed. Advanced mode uses an advertiser-specific model with more detailed modelling.

Google also recommends sending cookieless pings when ad_storage or analytics_storage is denied. Modelling may still be available if they are blocked, but Google then cannot generate advertiser-specific calibration factors, which may affect modelling accuracy. Google says these pings are never used to track individual users across apps or websites, build remarketing lists or generate user profiles.

The volume thresholds in Google Ads and GA4 differ

Google Ads conversion modelling has been available to eligible accounts since April 2021. It requires consent mode, or the IAB TCF v2.0 framework, to be correctly implemented. The account also needs 700 ad clicks over a 7-day period, per country and domain grouping. Google's help page calls this a "daily" threshold but describes the window as seven days.

Once those conditions are met, the models enter a training period and modelled conversions appear gradually. Google's developer documentation says the tag or SDK must meet a data collection threshold to protect user privacy.

GA4 behavioural modelling has its own criteria. Consent mode must run across all pages or app screens, and web properties must use the advanced implementation. The property needs at least 1,000 events a day with analytics_storage denied for at least 7 days. It also needs at least 1,000 daily users sending events with analytics_storage granted on at least 7 of the previous 28 days.

Google adds that meeting these figures does not guarantee eligibility. The model also considers factors such as the ratio of new to returning users and user-to-session counts. If a property loses eligibility, estimated data becomes unavailable, and it returns only from the date eligibility is regained.

How much of the gap modelling recovers

The Ads help page gives no single recovery rate. Its Ads help page gives one illustration: an advertiser with a 50% consent rate saw a 19% drop in conversions, and modelling produced an 18% conversion rate uplift. Google uses the example to show that conversion drops and modelled uplifts do not scale in step with consent rate.

The same page says consented users are typically 2 to 5 times more likely to convert than unconsented users. It adds that the ratio varies widely with consent rates, industry and conversion type.

Google's Ads documentation says the estimates use observable data from consenting users and historical trends, and aim to minimise over-prediction. GA4's version trains on observed data from the same property.

What stays unmeasured

Google states that it cannot observe whether a conversion by an unconsented user followed an ad interaction. Some conversions that in reality occurred may not be accounted for, as they are unattributable to ad-clicks without cookies.

GA4 limits modelling when consented traffic is thin. The documentation says behavioural modelling is applied only with high confidence in model quality. Without enough consented traffic, events from users who decline are not reported.

Within GA4, modelled data applies to metrics such as users, sessions and new users. Event counts such as page_view, first_visit and session_start are not adjusted in standard reports. Unconsented events become reportable there, but are not modelled.

Several features do not use modelled data at all: audiences, user explorer, cohort and user lifetime explorations, segments with a sequence, retention reports, predictive metrics and data export such as BigQuery. Estimated data may also not yet be available for the previous day.

A CAC figure built from the Conversions column blends observed and estimated conversions.

Reading modelled conversions when judging CAC and payback

In Google Ads, modelled conversions are added to the Conversions column at the same granularity as observed ones. Modelled conversion values appear in the Conversion value column when values are tracked. Every report using those columns is affected, as is bidding.

A CAC figure built from that column therefore blends observed and estimated conversions. In GA4, data export such as BigQuery and retention reports do not use modelled data, and neither do cohort explorations.

The thresholds apply per country and domain grouping, so one market can qualify while another does not. A single blended CAC across markets can then mix modelled and unmodelled data.

On bidding, Google's Ads help page says it recommended that advertisers who had switched from Target CPA or Target ROAS to Maximise conversions or Maximise conversion value to manage consent changes switch back after the modelling launch. Advertisers who changed targets should watch spend and return gradually to their previous ROI targets.

What to change first

The first check is the implementation type. On basic mode, Google Ads modelling relies on a general model, which is less detailed, and blocking cookieless pings may affect modelling accuracy. GA4 behavioural modelling requires the advanced implementation for web properties.

The second check is volume by market. Accounts below 700 ad clicks over 7 days in a country and domain grouping fall outside Ads modelling. GA4 properties below the 1,000-event and 1,000-user figures fall outside behavioural modelling.