Google Analytics 4: What It Actually Does (And Why It's Not What You Think)

When Google shut down Universal Analytics in July 2023, I watched a thousand marketers and analysts panic in real time. They didn't understand GA4. And honestly? That's because GA4 isn't an upgrade to UA — it's a completely different animal.

For years, UA had trained us to think about analytics in a very specific way: sessions, pageviews, bounces. It was a web-centric model built for a simpler internet. GA4 threw that out the window. It's event-based, mobile-first, and built around the idea that users move between your website, your app, and everywhere in between. Your job is to stop thinking like a session counter and start thinking like someone tracking user behavior.

The good news? Once you understand how GA4 actually thinks, it becomes a powerful tool for finding insights that matter. The bad news? Almost nobody knows how to use it properly.

This guide is built around the conceptual shifts you need to make to actually understand GA4, not just poke around in it. By the end, you'll know what GA4 is actually measuring, how to set it up so it measures the right things, and how to find insights that would take forever in the standard reports.

The Thing That Makes GA4 Different: It's Event-Based

Here's the core difference: Universal Analytics was session-based. A user arrived, did some stuff, and left. GA4 is event-based. Everything is an event.

A user lands on a page? Event. They click a button? Event. They watch a video, scroll past the fold, download a file, buy something — events all the way down.

This seems like a simple distinction, but it fundamentally changes how analytics works. In UA, you had to bend your data to fit into the session model. A user coming back to your site after three days? That's a new session. Did they come back because of an email campaign? UA would know if they clicked the campaign link, but the connection between "email sent" and "purchase 72 hours later" required complicated setup.

GA4 ditches sessions as the primary unit of analysis. Instead, it builds a timeline of events for each user (identified by a User-ID or a Browser ID). That timeline doesn't reset every 30 minutes. It's continuous. A user can have hundreds or thousands of events over months, and GA4 remembers all of it (well, keeps it for 14 months by default).

The enhanced measurement features in GA4 automatically capture common events without any configuration. Page views, scrolls, clicks on outbound links, file downloads, video engagement — GA4 is watching these out of the box. But here's the important part: automatic collection covers maybe 30% of what you actually need to understand your business. GA4 collects the friction-free stuff. It doesn't collect whether someone filled out a form, enrolled in a webinar, or became a paying customer.

That's on you.

What GA4 Collects Automatically (And What It Misses)

Let me be honest about the limits of auto-collection, because I've seen too many people assume GA4 just works.

GA4's enhanced measurement gives you:

Page views (tracked by page and page title)
Scroll engagement (when someone scrolls past 90% of the page)
Outbound clicks (links to external sites)
Site search (if you tag your search box properly)
Video engagement (YouTube videos embedded on your site)
File downloads (.pdf, .xls, etc.)
Form interactions (submits and interactions with form fields)

That's genuinely useful. But it's also incomplete.

Auto-collection won't tell you:

When someone completes a form but doesn't submit it (form abandonment)
Which product categories customers look at before buying
Whether someone watched your entire product demo video or bailed after 30 seconds
When someone adds something to a cart but leaves without checking out
Which pricing plan option gets the most clicks
Whether a user came back because of a specific marketing campaign (without proper UTM parameters)

The gap between what auto-collects and what you actually need to measure is where most teams stumble. They see GA4 tracking page views automatically and assume it's tracking "engagement." It's not. It's tracking navigation.

You need to set up custom events to capture the behavior that actually matters to your business. This is where the mental model really matters. In UA, you'd set up "goals." In GA4, you set up events and then decide which ones are conversions.

That's a crucial difference. An event is just a record of something happening. A conversion is an event you've marked as important to your business.

Setting Up Conversions That Actually Mean Something

Here's where I see teams go wrong: they convert every event into a conversion. Someone clicks the "Schedule a Demo" button? Conversion. Someone views the pricing page? Conversion. Someone watches a 3-second video autoplay? Conversion.

This is analytically useless. Your conversion count becomes meaningless. GA4 will happily report that you had 50,000 conversions, but if half of those are accidental clicks and page views, the number tells you nothing.

Conversions in GA4 should represent business outcomes. Real actions that move the needle. For a SaaS company, that might be:

Form submission (lead capture)
Account signup (user acquisition)
Trial activation (engagement)
Payment completed (revenue)

For an e-commerce store, it's probably:

Add to cart (if you want to track interest)
Purchase (revenue)
Newsletter signup (audience building)

For a media site, it might be:

Article read (content consumption)
Newsletter subscription (loyalty)
Membership signup (revenue)

The key is intentionality. Don't mark something as a conversion because GA4 makes it easy. Mark it as a conversion because it matters to your business and you want to optimize for it.

To set up a conversion in GA4, you go to Admin → Conversions and mark an event as a conversion. It's straightforward. But many teams never actually do this because they're overwhelmed by the decision. Start with the conversions that directly drive revenue or growth. You can add more later.

One more practical note: GA4's data-driven attribution requires a minimum of 400 conversions in the past 30 days to activate. If you're small and don't hit that threshold, GA4 falls back to last-click attribution (the event right before conversion gets 100% credit). This is important to know because a lot of teams try to turn on data-driven attribution and nothing happens — GA4 just silently uses last-click instead.

The Reports Nobody Uses But Everyone Should

Here's a frustrating truth about GA4: the default reports are mediocre. They're designed to be accessible, not insightful. They show you overall trends. They don't show you why.

That's where Explorations come in.

Explorations are GA4's custom reporting tool. You build them from scratch, picking dimensions (the "what" — page, user location, device, campaign, etc.) and metrics (the "how much" — page views, conversions, revenue, etc.). You can add up to 10 dimensions and 10 metrics per exploration.

There are a few exploration templates that are worth learning:

Funnel Explorations show you drop-off. You define a series of events (view product → add to cart → view cart → checkout → purchase) and GA4 shows you where users bail out. This is gold. An e-commerce team that looks at funnel data once a month learns more about their business than one that looks at the dashboard every day.

Path Explorations show the journey users take. You can see: "People who hit the pricing page and then the features page are 2x more likely to convert than people who skip pricing." This type of pattern is invisible in standard reports.

Segment Overlap lets you compare user segments. "What percentage of users who clicked on the spring campaign also visited the help center in the same week?" Overlap tells you if different acquisition channels are reaching the same people or different people.

Cohort Explorations let you track behavior over time for groups of users. "Users who signed up in January — how are they behaving now? Are they still logging in? Are they inactive?" Essential for understanding retention and lifecycle.

Learning to build explorations takes maybe 30 minutes. But it's the difference between understanding your data and just reading numbers.

Attribution in GA4: Which Model to Trust

Attribution is the problem GA4 was actually built to solve.

In the old UA world, you had a simple rule: give 100% of the credit to the last touch point before conversion. User clicks an ad, browses your site for 20 minutes, then buys something? The ad gets 100% credit. Easy. Also completely wrong.

GA4's data-driven attribution tries to fix this by using machine learning to figure out which touchpoints actually contributed to the conversion. Instead of arbitrary rules (first click, last click, linear), data-driven attribution looks at your actual data and says: "Based on historical patterns, this particular user was influenced by these marketing channels."

It's better than last-click. It's not perfect, but it's better.

The catch? You need 400+ conversions in 30 days for it to work. Below that threshold, GA4 uses last-click attribution automatically. If you're a B2B company with 50 deals a month, that's a problem. You might not hit 400 conversions for months.

There are other attribution models GA4 offers: first-click (gives credit to the first touchpoint), linear (splits credit equally across all touchpoints), time-decay (gives more credit to recent touchpoints). Each has different use cases.

Here's my actual advice: look at all of them. GA4's Attribution Report lets you toggle between models to see how the story changes. When you do:

Last-click makes your most recent channel look amazing
First-click makes your brand and top-of-funnel channels look amazing
Data-driven usually lands somewhere in between

If your channels all tell roughly the same story across models, you can trust the story. If they diverge wildly, you have an attribution problem that no model solves — usually because your customer journey is too complex or your tracking is incomplete.

The Features That Are Actually AI-Powered

GA4 ships with some AI features, and I need to be honest about what works and what doesn't.

Predictive audiences use machine learning to identify users most likely to purchase or to churn. In theory, this is gold — you could automatically add high-purchase-intent users to a remarketing audience. In practice, it's mediocre. The predictions work reasonably well if you have tons of conversion data and clear patterns. If your business is atypical or your conversions are sparse, the predictions are basically guesses. I'd test them, but I wouldn't stake my whole strategy on them.

Anomaly detection highlights unusual activity. A sudden spike or drop in pageviews triggers an alert. This is genuinely useful for spotting problems (traffic crashed? Discover why) but less useful for spotting opportunities (traffic spiked 20%? Could be anything).

Automated insights surface noteworthy changes. You open GA4 and it says "Conversions are up 15% this week" or "Users from Canada are converting 2x better than last month." These are good for spotting trends quickly, but they're not a replacement for actually exploring your data.

Here's what I'd say about GA4's AI features: they're helpful for busy teams. They'll flag things you should investigate. But they don't replace understanding your actual data. They're a starting point, not a conclusion.

Where Emilytics Fits In

Here's the thing about GA4 that I've seen over and over: teams set it up, turn on the auto-collection, mark a few conversions, and then ignore it. They look at the revenue number every month and call it a day.

GA4 is full of insight. But finding those insights requires:

Building multiple explorations to test different hypotheses
Comparing attribution models and deciding which tells the truth
Tracking down why metrics changed by cross-referencing three different reports
Exporting data, cleaning it, putting it in a spreadsheet to do the math
Coming back next month and doing it all again

This is fine if you're a dedicated analyst. It's painful if you're a founder or marketer trying to understand your business.

An AI agent like Emilytics connects to GA4 and does this work for you. You ask a question: "Why did conversions drop last week?" The agent pulls the relevant data, cross-references attribution, checks for anomalies, and tells you what happened. Not a chart. Not a raw data dump. An actual explanation.

This is where the mental model shift matters. GA4 gives you raw power — you can measure almost anything. But you have to actively explore to turn that power into insights. An AI agent automates the exploration. You get insights without spending hours in the interface.

FAQ

Can I use GA4 if I also use Google Ads?

Yes, and you should connect them. GA4 integrates with Google Ads automatically if you have a conversion pixel installed. This lets you see which ads drove conversions without relying entirely on GA4's attribution.

Is GA4 GDPR compliant?

GA4 can be GDPR compliant if you configure it properly: use anonymized IPs, don't send personally identifiable information, and get proper user consent. But GA4's default setup is not GDPR compliant. You have to configure it. Many European companies have struggled with this, so if you're in the EU, spend time on the legal setup.

Should I export GA4 data to BigQuery?

If you have the technical resources, yes. BigQuery import is free up to 10GB per month, which covers most businesses. BigQuery gives you raw event data that you can slice and dice in SQL. This is substantially more powerful than GA4's standard reports. If you have someone who knows SQL, this unlocks a lot of value.

What if I miss the UA migration deadline?

You missed it. UA is gone. Get GA4 set up immediately. If you still have UA data, export it before it's deleted. GA4 isn't a perfect replacement (it measures things differently), so historical comparison is always going to be messy, but it's better than having no data.

About Emily Redmond: Emily is a Data Analyst at Emilytics who spent three years watching teams struggle with GA4 after the UA shutdown. She's passionate about turning raw data into actual business insights and even more passionate about explaining why GA4's mental model matters. When she's not wrestling with event schemas, you'll find her thinking about the gap between what data shows and what people actually do.