Analytics tools record events. A page view, a click, a scroll past 75 percent, a session that ended after eleven seconds. What the person actually wanted is written down nowhere.
That gap is the whole job. Analyzing user behavior is not a tooling problem, it is a reconstruction problem: you have traces, and from those traces you try to work out what somebody was trying to do. It works better than most people expect, as long as you follow a few rules.
Site-wide numbers hide the thing you are looking for
A bookkeeping software company had an average engagement time of 38 seconds across the site. A number you can do absolutely nothing with.
Split by landing page, the same week looked like this. Visitors who arrived on the pricing page from a search for "invoicing software small business" stayed for just under three minutes. Visitors who arrived on the home page were gone after fourteen seconds. Two entirely different situations, blended into one figure that describes neither of them.
So the first rule: site-wide numbers are useful for noticing that something happened. To find out what happened, you need segments. Three of them almost always reveal something.
- Landing page. Somebody entering through a product page arrived with a different question than somebody landing on the home page.
- Source. Search visitors typed out their intent. Direct visitors already know who you are. Averaging the two together destroys both.
- Device. Forms that get filled in without a thought on a laptop get abandoned on a phone.
Three traces that tell you somebody got stuck
Frustration is visible in the data once you know its shape.
The reload. If the same session requests the same page again within a few seconds, somebody clicked something and nothing happened. A form submitted without feedback, a filter that did not apply, an image that never loaded. In GA4 you spot it as a second page_view for the same path inside one session.
The ping-pong. When visitors bounce between two pages three or four times, one of them is missing information they need side by side. Nine times out of ten it is the price sitting on a different page from the specification.
Internal search. The single most honest piece of feedback you will ever get, and the one most often left switched off. When people search your own site for "returns", "phone number" or "cancel subscription", that information is either missing or filed somewhere nobody thinks to look. Turn on search term capture and read the list once a month. It takes ten minutes and it will change at least one page.
Time on page for single-page visits was always fiction
Classic analytics calculated time on a page as the gap between two timestamps. If somebody views exactly one page, there is no second timestamp. The time on page is not short, it does not exist, and plenty of tools simply print a zero and move on.
GA4 measures this differently now, using the time the tab is actually visible in the foreground. That is much closer to reality. The old numbers and the old benchmarks are still circulating in reports, though. If your tool shows 0:00 for half your site, you have found a measurement artifact, not a behavior. How closely that ties into the metric everybody argues about is covered in understanding bounce rate.
Your numbers are lower than reality
Something many people never realize: the figures in your analytics do not describe all visitors, only the ones who consented to tracking. Every cookie banner someone dismisses or declines means a visitor who happened but appears in no report. Depending on your audience, a quarter to a half decline.
That has two consequences worth keeping in mind. First, your absolute numbers are too low, systematically. When GA4 shows a hundred visitors, the real figure might be a hundred and fifty. Second, and more important: as long as the decline rate stays reasonably stable, your comparisons still hold. Whether one page has twice the sessions of another is true even when both are undercounted. That is why fixating on absolute numbers is a mistake on small sites. Ratios and change over time tell you more than the bare counter, which only ever shows a slice anyway.
Four questions that carry an analysis
Any useful analysis answers four questions, in this order.
- Where do they arrive? Sort landing pages by sessions. On most sites the first ten account for more than eighty percent of everything. The tail is noise and it will eat your afternoon. Which GA4 report shows you this, and how to shape it, is in reading GA4 reports.
- What were they looking for? Pull the top queries per landing page out of Search Console. Then reread the page with that exact question in your head.
- Where do they stop? Exits at the end of a flow are not interesting. A person exiting on the thank-you page is a success. A person exiting at step two of four is a construction site.
- What do the others do differently? Build a segment of every session that converted and compare it against the rest. Usually exactly one thing differs: a specific page they saw, or the device they were on.
What to do tomorrow morning
Open a spreadsheet and give it half an hour. One row per landing page, the top ten are enough. Four columns: sessions, engaged session rate, top query from Search Console, and then one column you fill in by hand: what does the person arriving here want?
That last column is manual work and it is the valuable part. The moment you stall on a row and cannot write the sentence, you have found a page with no clear job. That is almost always the page where things go wrong, and often the one carrying a technical fault as well. Which faults to check first is covered in finding SEO problems.
Traffalyzer takes over the mechanical half of that exercise. It pulls landing pages, matching search queries and engagement figures into one view so you are not jumping between three tools with three different date ranges. The last column is still yours to write. Only somebody who knows the business can fill that one in.