CassandraVisit Website →
← Back to blog

Consent Mode Measurement: What You Keep, What Is Modeled

October 23, 2026
Consent Mode Measurement: What You Keep, What Is Modeled

Consent mode lets Google tags adjust their behavior to each visitor's cookie choice, and Google then uses modeling to estimate the conversions it can no longer observe directly. That keeps your reports usable. It does not tell you whether those conversions were caused by your ads. This guide explains what consent mode measures, what it models, the thresholds that decide whether modeling happens at all, and how to check the result against real incremental impact.

Cassandra
Measure Marketing Beyond Modeled Conversions
Cassandra analyzes market data and marketing channels to help you understand what drives sales and where your advertising investment has impact.
Explore Cassandra

TL;DR

  • Consent mode is not a consent banner. It passes the choice made in your banner to Google tags, which then change how they read and write cookies.
  • Advanced implementation loads tags on every page and sends cookieless pings, which is what feeds modeling. Basic blocks tags until consent, so there is nothing to model from.
  • Modeling has eligibility thresholds: 700 ad clicks over 7 days per country and domain grouping for Google Ads, and 1,000 daily events with analytics storage denied for GA4 behavioral modeling.
  • Modeled conversions are estimates of observed-versus-unobserved journeys. Validate them with holdout tests and a media mix model before you move budget.

Table of Contents

When a visitor declines ads or analytics cookies, consent mode, using Google AI, adjusts the behavior of the relevant tags so they do not read or write cookies for those purposes. Without cookies, you lose the link between an ad click and a later purchase for that visitor. Modeling is Google's attempt to fill that gap.

Two points are often confused:

  1. Consent mode does not collect consent. Your consent management platform or banner does. Consent mode is the channel that carries the visitor's choice to your tags.
  2. Consent mode does not restore the lost identifiers. It estimates aggregate outcomes. You will not recover a named user journey, only a modeled count.

The setup follows a fixed order. Google's developer guide has you set a default consent state before the visitor chooses, then update that state when the visitor interacts with your consent settings. A frequent failure is timing: consent updates must be sent well before the page unloads, because an unloading page can interrupt the transmission and leave sessions incompletely reported.

Basic versus Advanced: the choice that decides your data

This is the single most consequential setting for measurement.

BasicAdvanced
Tag behavior before consentGoogle tags are blocked and do not loadGoogle tags load on every page
Data sent when consent is deniedNoneCookieless pings without identifiers
Modeling inputOnly consented trafficConsented traffic plus cookieless signals
Typical riskLarge blind spots in reportingNeeds legal review of your own privacy position

Advanced consent mode depends on tags loading on every page load so that they can capture cookieless pings before any user interaction. Many brands block all tags until consent and, without realizing it, end up on the Basic path. They then miss the advertiser-level modeling that Advanced makes possible.

Whether Advanced is appropriate for you is a legal and privacy question as much as a technical one. Take it to your privacy counsel and your consent platform vendor, not only to your analytics team.

Version 2 is not a cosmetic update. It adds two parameters, ad_user_data and ad_personalization, alongside ad_storage and analytics_storage. Google's page on consent mode updates for traffic in the European Economic Area states that advertisers must adhere to the EU user consent policy to use ad personalization and for measurement with tags that send data to Google. If you serve EEA visitors on an older setup, you lose ad personalization and remarketing features for that traffic.

If your audience is mainly in the United States, this requirement still matters when you have any EEA traffic, and your US state privacy obligations are a separate question. Treat the EEA rule as specific to the EEA, not as a universal one.

GA4 users should also review the guidance on how to verify and update consent settings in Google Analytics, since Google has been strengthening enforcement of its EU user consent policy.

The thresholds that decide whether modeling happens

Modeling is not automatic. Google trains models on observed data, so small properties often get none.

To qualify, you need a minimum number of ad clicks in a 7-day window per country and domain grouping: 700 clicks. Once thresholds are met, a training period begins, and modeled uplift is reported in the Conversions column and flows into downstream reporting. If you run many small country-level campaigns, you can fall below the line in each grouping even when your total spend is meaningful.

GA4 behavioral modeling

For users who decline analytics cookies, GA4 requires enough data to train the model: at least 1,000 events per day with analytics_storage denied for at least 7 days, and at least 1,000 daily users sending events with analytics_storage granted for 7 of the previous 28 days. Even then, modeling is not guaranteed if the variance in the observed data is too high to train the models.

GA4 behavioral modeling eligibility thresholds

You can also review how much consent mode is changing your numbers in the consent mode impact results, which shows the effect of consent mode on your reporting.

Consent is not random. People who accept cookies tend to be more engaged, so they convert at a higher rate than people who decline. Google's documentation describes consented users as typically converting several times more often than unconsented ones, with the ratio varying by industry, conversion type and consent rate.

Here is an illustrative calculation, not a benchmark. Suppose consenting visitors convert at three times the rate of non-consenting ones, and half of your visitors consent. Out of every 100 visitors, consenters produce 1.5 units of conversion for every 0.5 produced by decliners. Your tags observe 75% of total conversions, not 50%. A 50% consent rate is a 25% observational gap here, and modeling aims to estimate the rest.

Two lessons follow:

  • Do not extrapolate from consent rate to lost revenue. The loss depends on the conversion ratio between groups, which differs for each business.
  • Do not assume modeled conversions are free of bias. They are estimates derived from the behavior of similar consenting users, and the consenting group is not a representative sample.

Consent mode answers one question: how many conversions probably happened that tags could not see. It does not answer the question budget owners care about: how many of those conversions would have happened without the ad.

Three gaps remain.

  1. Attribution is still last-click or data-driven within Google. Modeled conversions inherit the attribution logic of the platform that reports them.
  2. It covers Google tags only. Meta, TikTok, email and affiliate traffic each have their own consent handling and their own modeled numbers.
  3. Platform reports are self-reported. Each platform has an incentive to claim credit, and modeled uplift adds estimated credit on top.

This is where measurement approaches that do not depend on cookies earn their place. Our guides on cookieless attribution and privacy-first measurement with first-party data cover the data foundations. For an outside view on the broader shift, see this piece on analytics-driven ROI gains in modern attribution.

A practical stack has layers, each answering a different question.

  • Confirm that the default consent state is set before any Google tag fires.
  • Send the update as soon as the visitor chooses, not at page unload.
  • Decide on Basic or Advanced deliberately with legal input, and confirm tags actually load on every page if you choose Advanced.
  • Pass all four signals if you have EEA traffic.

Layer 2: Platform modeling as an input

Treat modeled conversions in Google Ads and GA4 as a better estimate of volume, not as proof of causation. Check that you clear the thresholds, and compare modeled uplift with your own backend orders by day.

Layer 3: Backend and first-party reconciliation

Your order system does not depend on a browser cookie. Compare orders, revenue and new customers from your own database with what platforms report. If platforms claim more than you sold, the overlap is the first thing to resolve.

Layer 4: Causal measurement

This is the layer that consent mode cannot supply. Incrementality testing withholds ads from a comparable group and measures the difference in outcomes, using aggregate results rather than individual tracking. Because it relies on totals from your own systems, consent choices do not erode it. Marketing mix modeling uses aggregate spend and sales over time and works without user-level identifiers at all.

A useful pattern is to calibrate: use experiment results to set or check what the mix model and the platform-modeled numbers say about each channel. If an experiment shows a channel's incremental contribution is half of what platform reports claim, adjust the budget on the experiment, not on the report.

A short validation plan for the next quarter

  1. Week 1: Audit consent flow timing, tag loading behavior and signal values. Record your consent rate by country and device.
  2. Week 2: Check threshold eligibility in Google Ads and GA4. Note which country groupings fall short.
  3. Weeks 3 to 4: Reconcile platform-reported and modeled conversions against backend orders by day and channel.
  4. Weeks 5 to 8: Run a geo or audience holdout on your largest paid channel. Our guide to testing incrementality without hurting live performance covers how to keep spend protected during a test.
  5. Ongoing: Feed results into budget decisions, and repeat tests where marginal returns look to be changing.

If you plan Q4 budgets, do this audit before you lock allocations. Consent loss tends to distort channel comparisons unevenly, because consent rates differ by region, device and audience.

We see consent mode as necessary plumbing, not a measurement strategy. It reduces data loss from one source, in one ecosystem, and it does so with a model whose assumptions you cannot inspect.

The risk is subtle. A modeled uplift raises reported conversions, ROAS looks healthier, and the temptation is to read that as performance. But a model that estimates unobserved journeys from observed ones will, by construction, follow the platform's own attribution. It can make a channel look better without showing that the channel caused anything.

Our view is that the answer is a loop. Plan the budget, run experiments that isolate real incremental impact, learn from the results, and apply them to the next allocation. AI agents can act quickly on platform signals, but they cannot prove that an action made money. Experiments and mix models can, and they do not depend on a cookie banner.

We would also encourage skepticism about any single number, including ours. Test the claim in your own context, with your own holdouts, and keep the confidence intervals in view.

— Gabriele Franco

Where Cassandra fits

If your team has fixed the consent setup and now needs to know which channels truly drive sales, Cassandra brings incrementality testing, marketing mix modeling and always-on incrementality together in one platform, so budget decisions rest on causal evidence rather than modeled platform credit. Plans on the pricing page start at €3,000 per month for Essentials, €3,750 per month for Complete and €5,400 per month for Scale, with Enterprise available on request.

FAQ

If you send data from EEA visitors to Google, Google's policy requires you to adhere to the EU user consent policy to use ad personalization and measurement, and v2 adds the two signals that carry that choice. Without it, you lose ad personalization and remarketing for that traffic. If you have no EEA traffic, the requirement does not apply in the same way, though you should confirm your obligations with privacy counsel.

No. The banner collects the visitor's choice. Consent mode passes that choice to Google tags so they behave accordingly.

Why is my modeled conversion uplift missing?

The most common causes are not meeting the volume thresholds, such as 700 ad clicks over 7 days per country and domain grouping in Google Ads, or running Basic mode, where tags are blocked before consent. GA4 may also skip modeling if the data variance is too high to train on. Check your impact results report first, then your tag loading behavior.

Can I measure marketing without cookies at all?

Yes. Incrementality tests and marketing mix models work from aggregate data such as spend, orders and revenue by period or region, so they do not need user-level identifiers. Many teams use them to validate what platform-modeled conversions claim.

Sources