How Google Consent Mode actually works (and why it still loses data).

Google Consent Mode v2 is often presented as the fix: “Don’t worry about users who decline cookies — Consent Mode will model the missing data for you.” But what does it actually do, how does the modeling work, and how much data is still lost?

What Consent Mode does at the network level

When a visitor lands on a site running GA4 with Consent Mode enabled, the Google tag checks the consent state before deciding what to send.

Consent granted (full mode)

GA4 operates normally: sets _ga and _gid cookies, sends full event payloads to google-analytics.com.

Consent denied (cookieless pings)

GA4 sends a stripped-down “ping” containing: a timestamp, the page URL, a session-scoped random value (not persistent), consent state flags, and basic browser metadata. These pings do not contain a persistent client ID, cookie values, user-scoped properties, or e-commerce transaction data.

How Google models the missing data

1. Collect consented data: GA4 has complete, cookie-based data for visitors who accepted.
2. Count cookieless pings: GA4 knows how many non-consented events occurred.
3. Build a behavioral model: Google uses consented visitors as a training set, assuming non-consented visitors behave similarly.
4. Scale up: The model multiplies consented data by a factor derived from the ping-to-cookie ratio.

Where the model breaks down

1. Selection bias

Users who accept cookies tend to be less privacy-conscious and more likely to engage with marketing. Users who decline are disproportionately tech-savvy professionals, privacy-focused browser users, returning visitors who habitually decline, and mobile users who quickly dismiss banners.

2. Small sample problem

Google requires minimum volume of consented data. For low-traffic sites, the model either produces no modeled data or highly volatile estimates.

3. No transaction-level accuracy

Modeled data cannot reconstruct individual transactions. A “modeled conversion” tells you nothing about what product was purchased, the order value, which campaign drove it, or the conversion path.

4. Ad blockers bypass Consent Mode entirely

Consent Mode only helps with visitors who see the consent banner and decline. It does nothing for visitors who block google-analytics.com at the browser or DNS level. The cookieless ping is never sent.

5. The “modeled” label is easy to miss

In the GA4 interface, modeled data is blended into reports with a small icon. Most users don’t realize the numbers they’re reading are partially estimated.

What this means in practice

• A/B tests based on modeled data may reach incorrect conclusions
• Attribution models are skewed for the non-consented majority
• Budget allocation decisions are unreliable for non-consented channels
• Compliance risk remains: several EU DPAs have questioned whether cookieless pings still constitute personal data processing

The alternative: remove the need for consent

The fundamental issue with Consent Mode is that it is a patch on an architecture that requires consent. Privacy-first analytics tools remove the need entirely by using first-party endpoints, discarding IP addresses, using session-scoped identifiers, and processing all data within the EU.

The result: 100% of visitors are measured with 100% real data. No modeling. No estimation. No consent banner required.