If you’ve been tracking website analytics the past few years, you’ve noticed something fundamental has shifted. The cookie era is fading. Privacy regulations are tightening. And your Google Analytics 4 (GA4) setup needs to change or miss out on vital data.
For marketers and companies that rely on GA4 for campaign tracking, attribution, and conversion insights, the upgrade isn’t just a compliance checkbox. It’s a rethink of how to measure performance with respect to user privacy.
So what is the solution then? GA4 has Consent Mode. It allows you to continue measuring performance even when users reject cookies, using AI to mathematically model the behavior of users who decide not to use cookies based on the data of those who opt in.
Let’s get into the mechanics of how Consent Mode actually works.
Table of Contents
- Introduction
- How does Google Consent Mode work?
- AI-Enabled Behavioral Modeling
- What changes post-implementation?
- The Future of Privacy-First Analytics
How Does Google Consent Mode Work?
Now, let’s look at the mechanics. What does Consent Mode v2 actually do when a user lands on your site and makes a consent choice?
Basics of Consent Mode v2
Consent Mode v2 is a framework that adjusts the behavior of Google tags according to the user’s consent. And this framework is commonly implemented through Google Tag Manager. When a visitor arrives at your site:
- Set default consent state to denied (prior to any tags loading)
- The user sees the Consent Management Platform (CMP) banner and makes a decision
- The CMP transmits consent signals using the Consent Mode API
- According to those signals, Google tags influence behavior
Each of the four consent parameters we saw before (analytics_storage, ad_storage, ad_user_data, ad_personalization) controls certain types of data collection.
Basic and Advanced Implementation
Consent Mode can be implemented in two ways:
| Implementation | How it works | Best for |
| Basic | Tags don’t fire at all without consensus | Simple sites, little tracking |
| Advance | Tags fire without consent with cookieless pings | Sites where accurate attribution is required |
Advanced mode is recommended because it sends anonymous activity signals (cookieless pings) even if users reject cookies. This data is used for behavior modeling purposes.
Tracking Without Identifiers: Cookieless Pings
That’s where Consent Mode gets clever. Even when analytics_storage is set to “denied,” Google still sends a cookieless ping, an anonymous request that includes
- Page URL
- Device type (mobile/desktop)
- Country (not a city)
- Type of Browser
- Time stamp
This ping does not include any personal identifiers (no IP address and no user ID). For users that refuse cookies, only the behavioral data is used for the model.
How Tags respond to consent status
Consent being given:
- GA4 uses first-party cookies
- Event tracking is working as expected
- The User-ID and client-ID are stored
- All the usual GA4 features are enabled
When permission is denied:
- No cookies are set.
- Only send cookieless pings
- Non-Identified Event Logging
- The data is marked “modeled” in reports
wait_for_update Parameter
wait_for_update is an important configuration in Consent Mode. This tells Google tags how long (in milliseconds) to wait for consent signals before firing:
| javascriptgtag(‘consent’, ‘default’, { ‘ad_storage’: ‘denied’, ‘analytics_storage’: ‘denied’, ‘wait_for_update’: 500 // Wait 500ms for CMP to update consent}); |
An interval set to 500 ms is recommended. Too short and you lose the signals of consent. Too long and tags load slowly, which impacts page performance.
What Conservation Modeling Needs
For behavioral modeling to work, your site needs:
- More than 1,000 daily events with analytics_storage=’denied’ for 7 consecutive days
- 1,000+ daily users with analytics_storage=’granted’ for 7+ of past 28 days
If you don’t have these thresholds, you won’t have modeled data available, and you’ll see gaps in your reports.
AI-Enabled Behavioral Modeling
This is where Consent Mode becomes really powerful. When users decline cookies, you don’t lose all their data; Google’s AI applies behavioral modeling to statistically estimate what their behavior would have been, based on users who opted in.
How Machine Learning Bridges Data Gaps
Here’s the short: If users agree to cookies, you get full behavioral data. Declined users just give you anonymous, cookieless pings. AI narrows the gap by identifying trends between the two groups. AI does this by applying predictive analytics to the estimate that is likely the conversion behavior.
For example: Say we have consenting users from Germany that are visiting on mobile between 6-8 pm and typically convert at 5%. Cookieless pings also contain non-consenting users from Germany that follow the same pattern. Now the AI estimates those non-consenting users have a conversion rate of about 5% as well.
Mathematical Basis
Behavioral modeling uses statistical inference to predict the behavior from observable signals. The model takes into account:
| Signal | Why this matters |
| Type of device | Mobile vs desktop users behave differently |
| Region/Country | Conversion rates are affected by geographic patterns |
| Time of day | Evening visitors may change differently than morning visitors |
| Browser type | Different behavior between Chrome and Safari users |
| Page URL | Results are influenced by landing page content |
| Origin of traffic | Organic search and paid ads behave differently |
The AI doesn’t know the identity of individual users. It uses aggregate patterns to estimate behavior at the group level.
Recovery Rate Conversion Model
When used correctly, Consent Mode v2 can recover up to 70% of the ad conversion data that would be lost if users opt out of cookies via behavioral data modeling.
That’s not 70% accuracy for each and every user. Which means:
- The overall counts of conversions are nearer to reality
- Attribution models (channel credits) are still more accurate. This improves reporting across the platforms like Google Ads.
- Although data are incomplete trend patterns are visible
How Google Checks for Accuracy
Google uses a holdback validation technique to keep modeling honest:
- A small group of consenting users is deliberately “denied.”
- Their real behavior is known (because they agreed)
- Model predictions are compared to real behavior
- The model readjusts if predictions do not match reality
This keeps the model calibrated and avoids biasing it systematically.
What is displayed in GA4 reports
When behavioral modeling is activated, GA4 tags data in two ways:
- Modeled metrics are marked with a data-quality icon (small triangle)
- Mixed reporting identity (observed + modeled data are mixed)
You will see metrics such as “Conversions (modeled)” and “Sessions (blended),” and you will understand what you’re looking at in terms of observed vs. estimated data.
Limitations of modelled data
The data being modeled is not perfect. Some GA4 features do not work with modeled data:
| Feature | Works with Modeled Data? |
| Standard reports (overview, acquisition) | Yes |
| Targeting for ads | No |
| User explorer reports | No |
| Forecasting metrics | No |
| Export raw data in BigQuery | No |
Most marketing teams are fine with standard reports. But if you want detailed user-level data, you still have holes.
Why is this important for your analytics
Without behavior modeling:
- You’d see 40-60% fewer European Economic Area (EEA) user sessions
- Conversion attribution would be heavily skewed towards consenting users
- ROI calculations on ad campaigns would be wrong.
Behavioral modeling:
- You get directional accuracy even with partial consent.
- Budget decisions are still data-driven, not guesswork
- You remain GDPR and DMA compliant without losing insights
What changes post-implementation?
After you’ve put Consent Mode v2 into practice, you’ll notice some differences in your GA4 reports. Knowing what to expect helps you not to panic when the data looks different and confirms that you are tracking correctly.
What Metrics Are Going To Get Better
If you have Consent Mode v2 enabled, you should see tangible improvements to the following key metrics:
| Metric | Expected to improve | Why does it change? |
| EEA meetings | +15-40% | Cookieless pings find previously blocked sessions |
| Conversions | +20-70% | Conversion modeling recovers lost attribution data |
| Clickthrough rates (ads) | +10-25% | Better conversion tracking improves calculation of CTR |
| Accuracy of attribution | Much better | Declining cookies no longer removes multi-channel paths |
These figures depend on your consent rate (percentage of users who agree). Modeling brings the biggest gains for sites with 30-40% opt-in rates.
How to find modeled data in GA4 reports
GA4 explicitly indicates that some data is modeled. Watch for these signs:
- The data-quality icon (small triangle) is shown next to modeled metrics in reports
- A blended reporting identity automatically applied (Combines observed and model data)
- More granular reports have labeled metrics such as “Conversions (modeled)” or “Sessions (blended).”
If you see these markers, you are looking at estimated data, not observed data. This transparency gives you a feeling of confidence.
What features fail to work with modeled data
Some features in GA4 will not work with modeled data. Here’s what is limited:
| Feature | Modeled Data Works With | Other options |
| Acquisition, Engagement) Standard reports | Yes | None needed |
| Custom Reports (Explorations) | Incomplete | Use for trends, not absolute numbers |
| Ad audiences | No | Use only consented users or observe the data |
| User Explorer reports | No | No user-level data for non-consenting users |
| Predictive measures | No | Required large observed data sets |
| Export raw data in BigQuery | No | Imported data not getting exported to BigQuery |
Most marketing teams don’t need anything more than standard reports. You will still be able to see acquisition, engagement, and conversion trends clearly.
How to know if a model is running
To check that the behavioral modeling is working:
- Admin -> Data Settings -> Collection of Data
- Verify that “Behavioral and conversion modeling” is set to ON
- Be sure you’re hitting the minimum thresholds:
- 7 days in a row of 1,000+ daily events with analytics_storage=’denied’
- > 1,000 daily users with analytics_storage=’granted’ for >= 7 out of 28 days
If you don’t meet these thresholds, modeling won’t switch on yet. This behavior is normal for new implementations; it takes 1-4 weeks to gather enough data.
How Your Reports Will Look
Here’s what to watch for in key reports:
- Acquisition Report:
- Channel performance still visible (organic, paid, direct, social)
- Traffic sources could be reporting a little higher numbers than before
- Direct traffic might increase (cookieless pings don’t always show referrers)
- Conversion Report:
- Total conversions include observed and modeled conversions
- Last-click attribution is less accurate (model fills gaps in user journeys)
- Assisted conversions are more visible (multi-channel paths are no longer invisible)
- Real-Time Report:
- Only shows observed data (no real-time modeling)
- Sessions from consenting users show up correctly
- Sessions from users who have not consented are only visible if cookieless pings are sent
When is data quality a problem?
Not every data dump is a problem. But beware of these red flags:
| Problems | What it means | What to do |
| >50% Sessions Drop After Implementation | The consent banner blocks tags too early | Check wait_for_update (should be 500 ms). |
| Conversions = 0 for users from the EEA | The model was not triggered or did not meet thresholds | Check thresholds met. Wait 2-4 weeks |
| All data appears as modeled | Consent banner not sending any ‘granted’ signals | GTM preview test CMP config validation |
| No data quality icons | GA4 settings does not allow for modeling | Enable behavioral modeling in Admin > Data Settings |
The Bottom Line: Directional Accuracy vs. Precision Perfection
Consent Mode is not going to give you the granularity of Universal Analytics. It does give us directional accuracy.
- And you know which channels are doing better
- You can still see improvement week over week
- Budget allocation decisions continue to be data-driven, not guesswork
- You remain GDPR & DMA compliant without blind analytics at all
That’s the best combination of privacy compliance and actionable insight for most businesses.
The Future of Privacy-First Analytics
Data privacy regulations aren’t going away. What’s clear is that Consent Mode in GA4 is not a workaround; it’s the framework that lets analytics survive in a cookieless world while respecting user choice.
The real power of Consent Mode lies in behavioral modeling. For users who opt out of cookies, AI can be used to estimate behavior to maintain directional accuracy for conversions, attribution, and campaign performance without sacrificing GDPR or DMA compliance.
The simple, practical formula is:
- Get consent right with a CMP.
- Send signals to Google using Consent Mode v2
- Let AI model the gaps where cookies are declined.
- Read GA4 reports with awareness of what’s observed vs. modeled.
For marketers, that means less guessing and more reliable decisions, even in a more privacy-restrictive environment. And for businesses, it means staying compliant without going blind on analytics.
All in all, AI-powered modeling is only going to get better. As machine learning continues to evolve, the accuracy of modeled data will continue to improve. And, as technology moves ahead, Consent Mode will be an ever-more powerful tool for privacy-first analytics.
