Definition
Google Analytics 4 is the generation of Google Analytics that replaced Universal Analytics. The help centre describes it as a property type that collects event-based data from both websites and apps, and that is the defining change: where the earlier model was built around page views and sessions with events bolted on, GA4 treats a page view, a scroll, a click, a purchase and an app screen view as the same kind of record, an event with a name and a set of parameters. Users and sessions are derived from those events rather than being the primary unit. The registry one-liner holds: it measures web and app interactions as events, and key events (the help centre's current term for what earlier versions called goals and GA4 itself first called conversions) mark the events the business counts as valuable. The GA4 hub covers the platform end to end; this page fixes the vocabulary.
How it works
A GA4 property holds one or more data streams: a web stream for a site, and iOS and Android streams
for apps. Each stream collects events, either through the Google tag (gtag.js) placed on the site, through
GTMTag management system., or through the Firebase SDK in an app. The help centre groups events into four kinds,
listed in the table below. Every event carries parameters, some automatic (page location, language, screen
resolution) and some set by the implementer, such as value, currency and items on a purchase.
Parameters that should appear in reports are registered as custom dimensions and metrics.
| Event kind | Who defines it | Examples |
|---|---|---|
| Automatically collected | GA4, on install | first_visit, session_start, user_engagement |
| Enhanced measurement | GA4, toggled in the web stream | scroll, outbound click, view_search_results, file_download |
| Recommended | Google names them; you implement them | purchase, add_to_cart, view_item, generate_lead, login |
| Custom | You name and implement them | Anything not covered above |
Marking an event as a key event does not change collection; it tells reports and linked Google Ads accounts to treat that event as an outcome. Campaign source and medium are read from UTMURL tags identifying traffic source. parameters and from auto-tagged Google Ads clicks, and credit for key events is assigned by the property's AttributionAssigning conversion credit to touchpoints. setting, which defaults to data-driven attribution. Standard reports (Realtime, Acquisition, Engagement, Monetisation, Retention, User) are built from the collected data; Explorations allow free-form analysis; the BigQuery export delivers the raw event rows for anything the interface cannot answer. Google documents retention limits for user-level data and thresholds that hide small rows in some reports; check the help centre for the current values rather than relying on a remembered figure.
In practice
Most of the work in a GA4 property is deciding which events to send, naming them consistently, and marking the right handful as key events. The events and parameters guide covers the naming decisions and the setup checklist walks through the configuration steps that are easy to miss: defining internal traffic, linking Google Ads and Search Console, setting data retention, and enabling the BigQuery export before the history is needed. A property that was set up by accepting every default will report something, but rarely the thing the business wanted to know.
Key events for a furniture retailer
A furniture retailer sends the recommended ecommerce events from its site through Tag Manager: view_item,
add_to_cart, begin_checkout and purchase, each carrying an items array with item_id, item_name,
price and quantity. It marks only purchase and generate_lead (a showroom appointment form) as key
events. In a month the property records 48,200 view_item events, 6,100 add_to_cart, 2,350 begin_checkout
and 870 purchase events worth £312,000, plus 140 appointment leads. Because add_to_cart is not a key event,
the linked Google Ads account bids toward purchases and appointments, not toward basket adds, and the
Monetisation report shows the drop between checkout starts and purchases (about 63%) that the UX team takes as
its brief.
What it is not
GA4 is not Universal Analytics with a new interface. Sessions, bounce rate and goals are defined differently or replaced, so figures from the two models are not comparable even for the same site and period. It is not a tag manager: the Google tag or GTMTag management system. delivers the data, and GA4 only receives it. It is not an ad server or an attribution product in its own right; its attribution reports credit the sources recorded in its own data, which is a narrower view than a platform-neutral measurement programme. It is not Google Ads conversion tracking either, although key events can be imported into Google Ads and used for bidding. And it is not a complete record of visitors: consent choices, ad blockers and modelling mean the reported numbers are an estimate of behaviour, which Google's documentation describes rather than conceals.
Related terms
Three terms sit closest to GA4. GTMTag management system. is the usual way events reach it on the web, and the data layer conventions that Tag Manager relies on determine how clean the GA4 event stream is. UTMURL tags identifying traffic source. parameters are how GA4 learns the source, medium and campaign of a visit that did not come through an auto-tagged Google click, and inconsistent tagging shows up directly in the Acquisition reports as fragmented channels. AttributionAssigning conversion credit to touchpoints. is the rule set GA4 applies when more than one source touched a user before a key event; the model chosen in the property changes every acquisition figure, which is why reports should state the model alongside the numbers. Google Ads, Search Console and BigQuery are the products it links to most often.
Common mistakes
- Marking every event as a key event. The reports and linked ad accounts then optimise toward low-value actions; keep key events to the outcomes the business would pay for.
- Inventing event names that Google already recommends.
buyinstead ofpurchaseloses the built-in ecommerce reports. - Comparing against Universal Analytics history. The models differ; treat the change-over as a new baseline.
- Forgetting the BigQuery export. It is not retroactive, so the raw data starts on the day it is switched on.
- Leaving internal traffic and referral exclusions unset. Payment gateways and staff visits distort acquisition and conversion figures.
- Reading thresholded or sampled reports as exact. Small segments can be hidden or modelled; the help centre explains the icon that flags it.
- Registering parameters after the fact. A custom dimension only reports from the day it is created; define the parameters that matter before the campaign starts.
