EMPIREYO

← Knowledge base

Online marketing

GA4 done right: measure first, decide after

A dashboard full of figures inspires confidence, even when the figures are wrong. A tag that loads twice, your own visits being counted or an enquiry that never shows up as an event: you only notice when you start comparing. This article goes through the checks that turn GA4 into a measurement you can make decisions on.

 ·  by  ·  Reading time 13 min  ·  Online marketing  ·  updated 12 November 2026

In short

GA4 is right when it counts what you mean, and nothing twice. Check five things: does the script only load after consent, does every page come in once, is your own traffic filtered out, is your most important action a key event, and does that number match your inbox. Only then look at channels and attribution.

Are you measuring this yet?

WhatWhy it countsHow to check
Consent Mode v2Without correct consent signals you measure too little and not cleanlyCheck in GA4 whether consent signals come in after a consent choice
Server-side measurementBrowser measurement leaks through blockers and limitsCheck whether your tags run through an own server container
Conversions returned in fullAd algorithms learn on what you send backCompare platform conversions with your own records
Attribution settingLast-click punishes channels that work early in the journeyView your attribution model and compare channels in both views
Own key eventsStandard events say little about your funnelWalk your request route: is every step measured as an event?

The table above lists five points for advertisers: consent, server-side tagging, sending conversions back, attribution and your own events. This article goes one step earlier. Before you compare channels or move budget, the foundation has to be right: what does GA4 count, and does it count it correctly?

Everything here about how GA4 works comes from Google Analytics' own help pages, consulted on 4 November 2026. GA4 changes regularly; menu names in your account may differ slightly.

What does it mean for GA4 to be right?

GA4 will never count every single visitor. Anyone who does not consent should not be in it, and some browsers block measurement scripts. Being right means something else here: GA4 counts what you think it counts, nothing twice, without your own visits, and the direction of the figures matches what actually happens in your business.

A measurement that misses a steady share every month is usable: you can see whether things are getting better or worse. A measurement that counts double one month and not the next is dangerous: you see growth that is not there.

Which terms do you need to know?

TermWhat it is
EventEvery action GA4 records, such as a page view, a click or a submitted form
Key eventAn event you mark as important to your business, such as an enquiry
ConversionAt Google now: an action you use to measure and steer ads in Google Ads
Attribution modelThe rule that decides which channel gets the credit for a key event
Data retentionHow long GA4 keeps user-level data before it is deleted
Reporting identityHow GA4 recognises visitors; with Blended it can show estimated figures

Note the difference between key event and conversion. Google has separated the two: important events in Analytics are called key events, and conversion is now the word for what you use in Google Ads to measure and steer campaigns. Older conversion data is still available as key event data.

Check 1: does the script only load after consent?

GA4's cookies are not functional cookies. The Dutch Data Protection Authority (AP) also does not list Google Analytics among its examples of analytics tools that do not track visitors. So assume you need consent in advance; that is how we do it ourselves. Google describes two forms of Consent Mode. With basic, the tags only load after a choice in the banner, and with a no, nothing goes to Google. With advanced, the tags always load and send measurements without cookies when there is no consent. What the rules say about this is in our article on cookieless measurement.

Here is how to test it. Open your site in a private window, reject the cookies and look in your browser's developer tools, on the Network tab, for requests to google-analytics.com. With basic there should be none. Then click accept and see whether measurement starts. Note: according to Google, DebugView, GA4's testing screen, does not show events from visitors who did not consent. So always test there with a yes.

Check 2: does GA4 count every page once?

GA4 records a page view automatically every time a page loads. According to Google you cannot switch that off. A setting under Enhanced measurement decides whether a change in the browser history also counts as a page view. Sites that change page without reloading need that, but on other sites it can add extra views.

Double counting can happen because GA4 is on the site in two ways, for example in the site's code and also through Google Tag Manager, or through a plugin and a theme at the same time. Open DebugView with a test visit and click through three pages. If you see one page_view per page, it is fine. If you see two, find the second source.

Check 3: is your own traffic filtered out?

You and your colleagues visit your own site often. Those visits distort the figures, especially on a small site. GA4 can recognise internal traffic by your office's IP address. You first mark that traffic, then switch on a data filter that excludes it. Google warns that an active data filter is permanent: whatever it excludes never comes back. You can create up to ten filters per property.

Also look at unwanted referrals. If a customer returns from another company's payment page, GA4 can count that payment service as the source instead of the channel the customer came from. In your data stream settings you can list such domains, up to fifty per data stream.

Check 4: do you measure the action that earns money?

With Enhanced measurement, GA4 records a number of events without extra code. They are handy, but say little about your own business:

EventWhen
page_viewOn every page view
scrollThe first time someone gets close to the bottom of a page
clickOn a click on a link to another website
view_search_resultsOn a search results page on your own site
video_start, video_progress, video_completeOn embedded YouTube videos
file_downloadOn a click on a link to a file, such as a PDF
form_start, form_submitWhen someone starts filling in and submits a form

A submitted form is not yet an enquiry received. form_submit counts the moment of sending; whether the enquiry actually arrived, GA4 does not know from that. So rather measure the moment the confirmation appears, or the thank-you page. For a completed form or a request for information, Google recommends the event generate_lead. Using that name also fills the report for leads.

Then mark the event that really matters as a key event: the enquiry, the quotation request or the order. Not the scroll or the click. Keep that list short: every extra key event makes the reports harder to read.

Check 5: does the number match your records?

This is the most important check, and the simplest. Take a full month. Count the key event for enquiries in GA4. Count in your inbox or system how many enquiries actually arrived. Put them side by side.

  • GA4 counts fewer. That is normal: not everyone consents. As long as the difference stays about the same from month to month, you can steer on the direction.
  • GA4 counts more. Then something is wrong: a duplicate tag, an event that counts again on reload, or test enquiries. Find out before you decide anything.
  • The difference jumps. Then something changed on the site, in the banner or in the tag. Find out what and when.

Which attribution model do you choose?

A customer often visits more than once before making an enquiry, through different channels. The attribution model decides who gets the credit. GA4 still has three: data-driven, paid and organic last click, and Google paid channels last click. The first click, linear, time decay and position-based models were removed in November 2023.

Data-driven spreads the credit across several touchpoints, based on your own account's data. That is why you sometimes see half key events per channel. Last click gives everything to the last channel the customer clicked. According to Google, direct visits receive no credit in any model, unless the whole path consisted of direct visits.

Two practical points. In the Attribution models report, compare how a channel looks in both models before you move budget. If the conclusion changes with the model, the model is steering your judgement. And know that the chosen model applies to event-scoped reports; session-scoped dimensions, such as Session source, do not change with it. The lookback window is 90 days by default for most key events.

How long does GA4 keep your data?

For a standard property you choose a retention period of two or fourteen months for user-level data and key events. Age, gender and interest data is always kept for two months. According to Google, the period applies to explorations and funnel reports; the standard reports with aggregated figures stay.

On IP addresses, Google writes that GA4 does not store those of users in the EU. It only uses them to derive a rough location, such as city and country, and then discards them. You can also switch off the collection of more granular location and device data per region. State the retention period in your cookie or privacy statement; the AP asks you to tell visitors how long cookies stay active.

Why do you sometimes see estimated or hidden figures?

Two features can make your figures different from what you expect. With the Blended reporting identity, GA4 can estimate the behaviour of visitors without consent, but only with advanced Consent Mode and enough traffic: according to Google, among other things a thousand users a day with consent. A data-quality icon at the top of the report shows whether estimated figures are included.

GA4 also applies thresholds. With demographic data or search terms it can leave out rows when there are too few users, so that nobody can identify individual visitors. You cannot switch those thresholds off. According to Google, choosing a longer period can help.

What do you record, and who may do what?

A measurement only stays right if someone looks after it. Three things help:

  1. A one-page measurement plan. For each key event: what counts, on which page, since when and how you test it. Anyone who later changes the site then knows what not to break.
  2. A change log. Note the date of every new banner, new tag, new form or site move. If a figure jumps, you can see straight away what changed that week.
  3. Clear permissions. According to Google, in GA4 you need the Editor role to change, for example, data retention or filters, and at least the Marketer role for the attribution settings. Give those rights to as few people as possible, and make sure the account belongs to your business, not to an agency or a former employee.

Also connect Search Console to your site. Search Console shows how you are found in Google, including in the AI features, with Google's own figures and regardless of your cookie banner. Google describes how to put that data next to Analytics data. That way you see both how people find you and what they do next.

Finally, set one fixed moment a month for check 5: the comparison with your inbox. It takes little time, and it is the quickest way to notice that something has changed.

When this does not apply

If you do not advertise and have few visitors, GA4 may be more than you need. A privacy-friendly analytics tool or simply counting what comes in can be enough; see our article on cookieless measurement. For apps and for the paid version, Analytics 360, some settings differ from those described here.

If you run campaigns in Google Ads, another layer is added: conversions, the link between Ads and Analytics, and the counting settings. Those fall outside this article. You can read more about marketing that pays for itself in our article on marketing.

What we do not know: how many GA4 installations measure incorrectly. We have not researched that, so we do not give a figure. We have also not measured how many visitors to our own site give consent.

Frequently asked questions

How do I know whether my GA4 counts double?

Open DebugView with a test visit, accept the cookies and click through a few pages. If every page comes in twice, GA4 is on the site twice. Also compare the number of enquiries in GA4 with your inbox each month: more in GA4 than in your inbox points to double counting.

Why does GA4 count fewer enquiries than my inbox?

Because not everyone consents, and some browsers block measurement scripts. That is normal. As long as the difference stays about the same each month, you can steer on the direction. If the difference jumps, find out what changed.

What is the difference between a key event and a conversion?

A key event is an event you mark as important in Analytics, such as an enquiry. A conversion is now, at Google, an action you use in Google Ads to measure and steer campaigns. You create a conversion from a key event.

Which attribution model should I choose?

GA4 still has three: data-driven and two forms of last click. Choose one and also compare channels in the other model before you move budget. If the conclusion changes with the model, be careful.

May I use GA4 without a cookie banner?

Assume not. GA4's cookies are not functional, and the Dutch Data Protection Authority does not list Google Analytics among its examples of analytics tools that do not track visitors. You can load the script only after a yes in the banner. If you want no banner at all, choose an analytics tool that does not track visitors.

Do you dare to trust your dashboard? We go through the checks on your GA4 and tell you per point what is there, what is missing and what that distorts. You can read more about our approach under online marketing. You work directly with Nawid, with a fixed team of five specialists.

Written with the help of AI, reviewed and edited by Nawid Rasa, who is responsible for the content. Sources are listed below the article. Spotted an error? Email hi@empireyo.com.

Get started

Do you dare trust your dashboard?

Let us run the five checks on your measurement. You hear per point what is in place, what is missing and what that skews.

Or message us directly

We usually reply within two working days. Prefer to call or email? That works too. How we handle your details is in our privacy statement.

Sources

Consulted on 4 November 2026.

Empireyo · Zwolle Open GA4 and answer one question: how many requests came in last month, and does that match your inbox? If you have to guess, you know where Monday begins.