GA4 Explained: Events, Key Events, Reports and Explorations

GA4's model in one picture: events inside sessions inside a user — and the key events are simply the events you told it matter.

What is GA4?

Google Analytics 4 is the current version of Google Analytics — the tag on your website that records who came, from where, what they did and whether they did the thing you wanted. It replaced Universal Analytics, which stopped processing data on 1 July 2023. If you searched "what is Google Analytics" in 2026, GA4 is the answer.

It is the collect layer of the stack we map in our marketing analytics guide; dashboards, attribution and lifetime value are all built on what it records. It is not Google Search Console, which shows what happened before the click — queries, impressions, position. GA4 starts at the click and covers every source, not just Google. You need both.

This page is not the setup walkthrough either. If you need the property, the data stream, the tag and the enhanced-measurement toggles, that lives in our GA4 setup guide. Here we assume GA4 is already collecting and explain what it is collecting.

How does GA4 think? Everything is an event

The one idea that unlocks GA4: there are no page views, clicks or goals as separate things. There are only events. A page view is an event called page_view. A scroll is scroll. A purchase is purchase. Every row GA4 holds is an event with a name, a timestamp and a bag of parameters — the details: page_view carries page_location; purchase carries value, currency and the items. Google's page on how GA4 differs from Universal Analytics puts it plainly: every hit is an event, identified by a name and its parameter values.

Events arrive in four ways:

• Automatically collected. session_start, first_visit, page_view — nothing to configure.
• Enhanced measurement. Data-stream toggles that add scrolls, outbound clicks, site search, video and file downloads without code. Coarse: "outbound click" does not know the link went to WhatsApp.
• Recommended. Names Google pre-defined — generate_lead, purchase, add_to_cart. GA4's reports and Google Ads understand them without extra work.
• Custom. Anything you invent — whatsapp_click, brochure_download. Usually sent through Google Tag Manager, which is why GTM and GA4 are set up together.

A worked example. A Pune coaching institute's site has a form, a "Chat on WhatsApp" button and a phone number. Out of the box GA4 sees three click events at best. After an hour in Tag Manager it sees form_submit, whatsapp_click and call_click, each with a course_name parameter. Only the second version answers a business question.

Users, sessions and engagement: what the numbers mean

Events roll up into two bigger units. A session is one visit, ending after 30 minutes of inactivity by default. A user is one browser or app instance (or a signed-in person, if you pass a user ID) that can have many sessions.

Three definitions trip up anyone who learnt on UA, all from Google's metric comparison page:

• Engaged session. Lasts 10 seconds or longer, has one or more key events, or has two or more page views. This is GA4's core quality signal.
• Bounce rate. The percentage of sessions that were not engaged. A visitor who read your pricing page for two minutes and left is engaged, not a bounce.
• Active users. The "Users" figure in most GA4 reports is active users, where UA showed total users. Comparing the two across the 2023 migration is comparing different things.

One more: "Session" dimensions describe the source of that visit; "First user" dimensions describe how the user was first acquired. Read the wrong one and you credit the wrong channel.

What are key events in GA4?

A key event is any event you have flagged as important to the business. Google's definition, from its key events help page: "an event that measures an action that's particularly important to the success of your business." Any event you collect can be marked as one with a toggle in Admin → Events.

If you remember "conversions" in GA4, this is the same thing: Google renamed them in 2024 so that "conversion" could mean one thing across its products. Its conversions-vs-key-events page draws the line: a key event lives in Analytics and appears in Analytics reports; a conversion is what Google Ads uses to measure and optimise a campaign. The path is Event → Key event → Google Ads conversion, and the last step needs a linked Google Ads account.

Practically: marking whatsapp_click as a key event lets you see it by channel in GA4. It does not tell Google Ads or Meta anything — for that you import it or fire the platform's own tag, the subject of our conversion tracking guide.

What to mark, for a typical Indian lead-gen site: form submit, WhatsApp click, call click, brochure download and — only if the site sells — purchase. Attach a value where you honestly can: if a dental clinic knows a booked consultation is worth roughly ₹1,500 in first-visit revenue, put that on form_submit and every report starts speaking in rupees. If you cannot defend a number, leave it blank.

Standard reports vs Explorations: which one answers your question?

Standard reports (Reports → Acquisition, Engagement, Monetisation, Retention, plus Advertising) are pre-aggregated tables Google keeps for every property. Explorations (the Explore tab) are blank canvases you build from event-level data.

Each standard report answers one question:

Explorations are for the questions no standard report is shaped for. Google lists the techniques on its Explorations help page: free form (a pivot table you design, up to five row dimensions, ten metrics and four segments), funnel (where people drop between "viewed course page" and "submitted form"), path (what they did after the pricing page), segment overlap, user explorer, cohort and user lifetime.

The trade-off: standard reports are fast and stable but limited to Google's dimensions; Explorations are flexible but subject to retention, sampling and thresholds.

How to read GA4 to answer "which channel brought the leads?"

The question every owner actually has. Once key events exist it takes five minutes in the Traffic acquisition report:

1. Open Reports → Acquisition → Traffic acquisition and set the date range to the last full month, compared with the previous period, so every number has a direction.
2. Leave the primary dimension on Session default channel group for the first pass. Organic Search, Paid Search, Paid Social, Direct, Email, Referral — the view a client understands.
3. Open the dropdown on the Key events column. It defaults to all key events added together. Pick the one you mean — form_submit or whatsapp_click — so a scroll milestone someone once marked as key does not inflate the count.
4. Switch the primary dimension to Session source / medium when a channel looks interesting. "Paid Social" becomes facebook / cpc. A large Direct or Unassigned row means links are going out untagged — fix that with the UTM builder.
5. Add Session campaign as a secondary dimension and sort by the key event column, not by sessions. Sessions are activity; key events are results.
6. Read the session key event rate, not just the count. 8,000 sessions and 40 leads is not better than 900 sessions and 30 leads. The rate tells you where the next ₹50,000 should go.
7. Sanity-check in Advertising → Attribution before acting. Flip from the default data-driven model to last click; if a channel only looks good under one model, it is an assist, not a closer.

When we run performance marketing for a client, this read happens every week before anyone opens the ad platforms' dashboards — which will each claim the same lead.

Data retention, sampling and thresholding: three settings that quietly change your numbers

Data retention controls how long GA4 keeps event-level data — what Explorations are built from. On a standard property the choices are 2 months or 14 months; windows up to 50 months are for paid GA4 360. The part people miss, from the same page: the setting does not affect standard aggregated reports, only Explorations and funnels. A property on 2 months still shows a year of Traffic acquisition — but a six-month funnel in Explore comes back empty. Check Admin → Data settings → Data retention; Google's page does not say which option a new property lands on, so do not assume it is 14.

Sampling. Standard reports come from pre-aggregated tables and are not sampled in normal use. Explorations query event-level data, and past 10 million events on a standard property GA4 uses a representative sample. A clinic site never hits that; a D2C brand running a year-long exploration over a Diwali sale will, and the icon at the top of the exploration is the only warning. Shorten the date range or move the question to BigQuery.

Thresholding. When a report includes demographic or interest data, or too few users sit in a row, GA4 withholds the row so nobody can infer an individual. The thresholds are system-defined and cannot be changed; the only relief is a longer date range. Rows that add to less than the total have usually been thresholded, not lost.

Google's reporting surfaces comparison sums it up: standard reports can collapse long tails into an "(other)" row, Explorations can be sampled, and the BigQuery export is neither.

What is the GA4 BigQuery export, and does a free property get it?

Yes. The BigQuery export sends every raw event into Google's data warehouse, and it is available on standard properties — in the UA era it was a paid-tier feature. A standard property can export up to 1 million events a day into the free BigQuery sandbox, with charges only past the sandbox limits; the streaming (intraday) export costs extra per gigabyte, so most small businesses use the daily export only.

Why bother if you are not a data team? The export is the only copy of your data that outlives the retention window, it is never sampled or thresholded, and it is what Looker Studio queries when a dashboard outgrows the GA4 connector. Turn it on the day the property is created — it does not backfill. First queries: our BigQuery for marketers guide.

GA4 vs Universal Analytics: a translation for people who remember UA

If your instincts were trained on UA, most still work, but the vocabulary moved:

• Hit types → events. Pageview, event and transaction hits are all events now; category, action and label become parameters.
• Goals → key events. Any event qualifies, not just a destination URL or a duration.
• Bounce rate → engagement rate, with bounce rate redefined as its inverse.
• Views → gone. Use data filters, comparisons and Exploration segments; there is no "test view" protecting the main one.
• Last-non-direct-click by default → data-driven by default. Channel credit will not match your UA memory.

Your UA history is gone — Google removed interface and API access in July 2024 — so any comparison reaching before mid-2023 compares two measurement systems. Say so in the report.

The GA4 mistakes we see most in Indian accounts

• Traffic, zero key events. Thousands of sessions and nothing marked. Every report is trivia until form submit, WhatsApp click and call click are key events.
• Everything marked as a key event. Scroll depth, video start and "view pricing" all flagged, so the lead count means nothing. Two to five real actions is the norm.

Where to go from here

Get the model right first: events with useful parameters, two to five honest key events, retention checked, the BigQuery export on. Then the weekly read above becomes a five-minute habit rather than a monthly argument, and the rest of the stack — Tag Manager, conversion tracking, a dashboard — has something true to build on.