Consent Mode GA4: What It Changes for Tracking

The call I get most often after a cookie banner goes live sounds the same every time. The line in Analytics fell off a cliff on a Tuesday, the owner wants to know who broke the site, and the honest answer is that nothing broke. The confirmation step that now sits between a visitor and your tags is doing exactly what it was installed to do.

I implement consent mode for clients and then have to explain the aftermath, so this post is the measurement side of that conversation. What the change does, the gap it leaves, and how to read your own reports once it is in place.

The short version: consent mode is a protocol that tells Google what your visitors agreed to. It is not a plug-in that keeps your numbers intact, and it will not give you back a clean before and after comparison.

What consent mode actually changes

A protocol between your banner and Google's tags

Your banner collects a choice. Consent mode is the mechanism that passes that choice through to the tags on the page, so they know whether they are allowed to store identifiers. Per Google Ads Help on consent mode, "Consent mode lets you communicate your users' cookie or app identifier consent status to Google."

That is the whole job. It does not decide anything, it does not replace your banner, and it does not change what you are allowed to collect. It reports a state.

Why it is not a banner and not a reporting feature

People assume the switch in their tag settings is the banner they bought, and it is not. Per the same Google reference, "Consent mode doesn't provide a consent banner or widget." The banner is a separate product or a theme feature, and the two have to be wired together correctly or the signals never leave the page.

It is also not a reporting feature in the sense owners hope for. Nothing in it recovers the visitors who said no.

Why your numbers drop when a banner goes live

Users who decline are still missing from observed data

This is the sentence I send to owners when they ask why the graph fell. Per Google Analytics Help on behavioural modelling for consent mode, "When you implement a consent banner for your website or app, Analytics will be missing data for users who decline consent."

The size of the drop depends on how your audience behaves, and it can be large. Anyone who has spent time on an Australian consumer site knows that a meaningful share of visitors click the option that closes the banner fastest.

Separating a real decline from a measurement gap

Two checks settle it in about ten minutes. First, does the drop line up exactly with the day the banner or consent change went live, and is it a step change rather than a slope? Tracking changes produce steps. Demand changes are usually uneven, and they hit some channels and query types harder than others.

Second, look at a source that does not depend on your consent state. Search Console clicks come from Google's own data rather than your tags, so if clicks hold steady while sessions fall, you are looking at a measurement gap rather than a demand collapse.

Third, look at the shape across channels. If paid traffic barely moves while direct and organic sessions fall away together, the tags on your landing pages are the likely difference, because visitors arriving from an ad tend to accept and visitors arriving from a search result often dismiss the banner without reading it. If every channel drops by a similar proportion on the same date, suspect the tag change rather than a sudden change in demand.

The two implementations and what each costs you

Basic: consent withheld and data never sent

In the basic setup, tags load once a visitor makes a choice, and before that point nothing is sent. It is simpler to configure and it is the stricter of the two. The cost is a hard gap: declines are invisible in every report, and there is no signal at all from the period before a choice is made.

Advanced: cookieless signals before consent is given

The advanced setup sends a cookieless ping before consent is granted, and it is what makes behavioural modelling possible later, because the property has to see denied consent activity to learn from it. It is still not a recovery of missing users. It gives Google a signal and, in some properties, a modelled estimate.

If you are not sure which implementation is live, ask for it in writing rather than guessing from the dashboard. The difference shows up directly in your reports, and the only reliable way to know which one you have is to test it on a page.

Whichever route is in place, verify it rather than trusting it. I still find sites where the tag manager fires on some templates and not others, so a third of the site reports as though every visitor agreed.

What behaviour modelling can and cannot recover

Google describes behavioural modelling as a machine learning method that estimates the behaviour of users who decline analytics cookies by looking at the behaviour of similar users who accept them, as set out on the same Analytics help page. Read that sentence carefully, because it says model, not recover.

The eligibility thresholds Google publishes

Modelling only enters the picture if your property clears specific published thresholds. One of them is stated plainly in Google's documentation: "The property collects at least 1,000 events per day with analytics_storage='denied' for at least 7 days." The prerequisites also include at least 1,000 daily users sending events with granted consent across a 28 day window, and consent mode enabled across all pages.

A low traffic site may never clear those thresholds, which is worth knowing before anyone promises you modelled reporting.

Why meeting the thresholds is not a guarantee

Google adds a caution of its own: meeting the external prerequisites does not guarantee eligibility, because the underlying model applies its own eligibility criteria and data thresholds to keep modelled data accurate. In practical terms, modelling is a possible improvement rather than a fix, and your reporting should not be built on the assumption that it will appear.

How to read reports that include modelled data

Showing or hiding modelled data in Analytics

Analytics gives you a control for this in the reporting interface, and it is worth knowing where it sits before a board meeting rather than during one. Turn modelled data on and the totals get closer to what your site probably saw. Turn it off and you are reading only confirmed, consented activity.

Neither view is wrong. Mixing them without saying which is which is the mistake.

Which metrics move and which stay stable

Modelled estimates apply to specific metrics rather than to everything in the property, so two reports on the same screen can be built on different underlying data. When a number is modelled, treat it as an estimate with a direction rather than a count you can invoice against. Anything used to make a spending decision should be labelled.

What consent mode does not do

It does not answer your privacy questions

Consent mode is a measurement protocol that communicates consent state to Google's tags. It is not a banner, it does not assess your obligations under privacy law, and it does not decide what you are allowed to collect. Those questions belong with someone qualified to answer them, and I am not that person. Treat this post as the measurement half of the problem only.

It does not restore a like for like comparison

If your reports show a step change on the day the banner launched, no amount of modelling makes the two periods directly comparable again. What you can do is make the break visible. Annotate the date in Analytics, mark it on your reporting timeline, and set a fresh comparison baseline from the change date forward rather than comparing this month with the month before it.

A practical month one checklist after a banner change

• Confirm the consent signal fires on every template, not just the homepage and blog posts
• Confirm the default denied state is set before the tags load, rather than after
• Check the banner actually blocks storage for visitors who decline
• Annotate the launch date in Analytics and in your reporting document
• Record whether your property meets the published modelling thresholds, with the numbers you checked
• Agree which reports are observed and which can include modelled estimates
• Set a new comparison baseline from the change date forward
• Tell anyone who reads the reports what happened and when, in one paragraph

That last item is the one people skip. Six months later, a sharp eyed stakeholder finds the step in the graph and asks why nobody mentioned it.

Where to get the implementation checked before you trust a report

Most consent mode problems I find are installation problems rather than platform problems: a tag that fires twice, a default state set after the tag loads, a template that was missed, a banner that offers no clear way to decline, or a modelled figure quietly blended into a headline number.

A short review should leave you with four things in writing: which implementation you are running, the date the change took effect, the templates where the signal is missing or firing twice, and a plain statement of whether your property currently clears the thresholds. With those four on one page, every report you read afterwards can be read honestly, and the whole exercise takes an afternoon rather than a project.

Before you make another spending decision on the strength of your analytics, it is worth having your consent mode and analytics setup checked against what your reports are claiming. If the numbers also drive where your budget goes, measuring return when part of the data is modelled is a sensible next step, and a marketing audit that tests whether you can trust your numbers will show you where the gaps are. You can also ask for a consent and tracking review and get a written list of what is firing and what is not.

Frequently asked questions

Did my traffic actually fall, or is this a measurement gap?

Check whether the drop lines up exactly with the day the banner or consent change went live, and whether the fall is across every channel at once. A tracking change usually produces a step change on one date, while a genuine demand change is usually uneven across channels and query types. Search Console clicks, which do not depend on your analytics tag, are a useful cross check on whether real demand moved.

Does consent mode keep my numbers complete?

No. Google states that once you implement a consent banner, Analytics will be missing data for users who decline consent, and that behavioural modelling attempts to model those users from the behaviour of similar users who do accept cookies. Modelling is an estimate shown inside Analytics, not recovered data, and it only becomes available if your property clears specific thresholds.

What does my property need before modelling can apply?

Google publishes explicit prerequisites, including that consent mode is enabled across all pages and that the property collects at least 1,000 events per day with analytics_storage set to denied for at least 7 days. There is also a requirement for at least 1,000 daily users sending events with granted consent across a 28 day window. A low traffic site may never clear those thresholds, which is worth knowing before you promise anyone modelled reporting.

If I meet the thresholds, will I get modelled data?

Not automatically. Google's own documentation says meeting the external prerequisites does not guarantee eligibility, because the model also applies its own data thresholds and will simply not train if the data is insufficient. Treat modelling as a possible improvement rather than a fix, and keep reporting on the observed figures as your default.

What should I do to make reporting readable after the change?

Annotate the date. Note the change on your reporting timeline, mark it in your analytics annotations, and set a new comparison baseline from the change date forward rather than comparing month on month across the break. Also state in the report which figures are observed and which include modelled estimates, so nobody treats a blended number as a raw count.

Is consent mode a compliance solution?

Consent mode is a measurement protocol that communicates consent state to Google's tags. It is not a banner, it does not assess your obligations under privacy law, and it does not decide what you are allowed to collect. Get your obligations assessed separately by someone qualified to do that, and treat this post as the measurement half of the problem only.