If you are still asking what is GA4, you are asking about the only version of Google Analytics that still collects data. Universal Analytics stopped processing hits on 1 July 2023 for standard properties, and the old interface and its historical data were switched off the following year. GA4 is not an upgrade to Universal Analytics in the way a new phone OS is an upgrade. It is a different measurement model wearing the same brand name.
That distinction matters more than any button-location guide. Most of the confusion around GA4 comes from people carrying Universal Analytics muscle memory into a tool that never had sessions at its core. Get the model right and the reports stop feeling hostile.
The one-sentence definition
GA4 is an event-based analytics platform: every interaction on your site or app — a page load, a scroll, a click, a purchase — is recorded as an event with parameters attached. There is no hit type. There is no separate pageview, event, transaction and social interaction schema. It is events all the way down, and everything you see in a report is an aggregation of those events.
Universal Analytics organised data around sessions, and events were a second-class citizen bolted on with category, action and label fields. GA4 inverted that. A pageview in GA4 is literally an event named page_view. A purchase is an event named purchase carrying an items array. Sessions still exist as a derived dimension, but they are calculated from a session_start event rather than being the atomic unit of measurement.
GA4 basics: the vocabulary that actually changed
Half the battle is translation. A short glossary saves hours of confusion:
- Key events — what Universal Analytics called goals and what GA4 itself called “conversions” until Google renamed them in 2024. In the GA4 interface you now mark an event as a key event. Google Ads still refers to the ones it imports as conversions, which is why you will see both words in the same screenshot.
- Engaged session — a session that lasted 10 seconds or longer, OR fired a key event, OR produced at least two pageviews or screenviews. Any one of the three qualifies.
- Engagement rate — engaged sessions divided by total sessions.
- Bounce rate — in GA4 this is simply the inverse of engagement rate. It is not the Universal Analytics metric, so a GA4 bounce rate of 38% is not comparable to a UA bounce rate of 38%.
- User — GA4 reports “total users” and “active users”, and the headline number in most reports is active users, not the everyone-who-ever-loaded-a-tag figure UA showed.
If someone hands you a year-on-year chart that splices UA data onto GA4 data, treat it as decoration. The metrics have different definitions on either side of the join.
What GA4 collects for free the moment you install it
Enhanced measurement is switched on by default for web data streams, and it automatically collects a set of interactions without any tag manager work: page views, scrolls (at 90% depth), outbound clicks, site search, video engagement for embedded YouTube players, file downloads, and form interactions. That is a meaningful amount of behavioural data for zero effort.
It is also the single most common source of inflated numbers. Teams migrate, do not realise enhanced measurement is on, and add their own GTM tags for scroll depth and outbound clicks — so every scroll fires twice, once as scroll and once as a custom event. Before you build any tracking plan, open Admin, find your web stream, and read the enhanced measurement toggles. Decide what GA4 handles and what you handle. Never both.
GA4 setup: what installation actually involves
Setting up GA4 is genuinely quick when nothing exotic is involved. The steps for a standard website:
- Create a GA4 property in Admin, then create a web data stream for your domain. GA4 hands you a measurement ID that starts with
G-. - Deploy the Google tag. Either paste the gtag.js snippet immediately after the opening
<head>tag on every page, or fire a Google Tag configuration tag through Google Tag Manager on all pages. Pick one method; running both on the same measurement ID double-counts everything. - Confirm data is arriving in DebugView or the Realtime report before you close the tab.
- Set data retention to 14 months (the default for event-level data is 2 months, which quietly breaks year-on-year exploration reports).
- Link Google Search Console and Google Ads if you use them.
- Define your key events, and only then start marking events.
CMS platforms often shortcut step 2. WordPress plugins, Shopify’s native integration and Wix all take a measurement ID directly. Those integrations are fine — just make sure you have not also hard-coded the same tag in your theme header, which is the classic cause of every metric doubling overnight.
How to transition to GA4 when you are late
Plenty of sites are still running a half-finished migration. If that is you, the honest sequence is: get clean baseline collection first, then rebuild reporting, then rebuild attribution. Trying to recreate your old Universal Analytics dashboard on day one is a trap, because several of those metrics no longer exist in a comparable form.
Your old UA data is gone from the interface, so if you did not export it before Google’s shutdown, work from whatever BigQuery exports, Looker Studio extracts or spreadsheet snapshots you kept. Going forward, GA4’s free BigQuery export is the closest thing to an insurance policy — it gives you raw event rows you own, and it costs nothing to switch on. Turn it on today even if nobody plans to query it for a year.
The GA4 quirks that generate support tickets
Three behaviours account for most “GA4 is broken” complaints, and none of them are bugs.
Processing delay. Standard properties take 24 to 48 hours to fully process data. Realtime and DebugView are live; standard reports are not. Judging yesterday’s traffic at 9am is judging incomplete data.
Data thresholding. When Google Signals is enabled and a report segment is small, GA4 withholds rows to prevent identifying individuals. You will see a small icon near the report title and inexplicably missing demographic or landing-page rows. Turning off Google Signals in reporting identity, or widening the date range, usually restores them.
Attribution is retroactive. GA4 defaults to data-driven attribution. Change the reporting attribution model and historical numbers in acquisition reports change too, because credit is recalculated rather than stamped at collection time. That is by design, and it surprises everyone the first time.
What GA4 does not tell you about SEO
GA4 is excellent at what happens after the click and close to useless before it. It does not know which keyword produced a visit, it does not know your ranking position, and Organic Search in GA4 is a channel bucket, not a keyword report. Search Console holds the query, impression, click and average-position data; GA4 holds the behaviour and revenue. You need both, and they will never agree perfectly because one counts clicks at Google’s end and the other counts sessions at yours. A 5-15% gap between Search Console clicks and GA4 organic sessions is normal.
This is exactly why SEO Rocket connects both Google Search Console and GA4 and treats them as ground truth for your own site, showing real clicks and sessions beside third-party index estimates with each source clearly labelled. Third-party volume and position figures are modelled estimates; Google’s own numbers for your property are not. SEO Rocket does not replace GA4, build custom GA4 reports for you, or implement tracking code on your site — you still own the tag. What it does is stop you from arguing with an estimate when the real number is sitting one panel over.
Where to start this week
Do four things in order. Audit for duplicate tags by loading a page with a tag assistant and counting how many times page_view fires. Read your enhanced measurement toggles and remove any manual tags that duplicate them. Push data retention to 14 months and switch on the BigQuery export. Then define the three or four key events that genuinely represent business outcomes, and ignore the rest.
GA4 rewards a small, deliberate measurement plan far more than an exhaustive one. Once collection is trustworthy, the reports stop being an argument and start being evidence.
