Web traffic analytics explained

Web traffic analytics is the narrowest and oldest part of this market: counting who arrived at a website, where they came from, what they read and whether they stayed. It is not product analytics, which follows identified people through a piece of software over weeks, and it is not behaviour analytics, which records what happened on a single page. Those are three different jobs and buying across them is the most expensive mistake in this market.

This page is the category explainer. What the metrics mean, which ones are defined inconsistently between vendors, why two tools on the same site will never give you the same number, and how to pick one once you understand all that.

What each metric actually counts

Four terms do most of the work, and three of them are defined differently by different vendors, which is the root of most reporting arguments.

Pageviews is the only one everybody agrees on: one view of one page. Reloads count. It is the least interesting number and the one most often quoted.

Visitors, sometimes called unique visitors or users, is where the definitions diverge. A cookie-based tool counts a browser it recognises, which means the same person on a phone and a laptop is two visitors, and a person who cleared their cookies is a new one. A cookieless tool counts a token derived per day, so the same person on two days is two visitors by design. Neither is wrong. They are answering different questions and the numbers are not comparable.

Sessions, or visits, is a group of activity from one visitor with a gap threshold, traditionally thirty minutes. Some tools have abandoned the concept entirely, which is a large part of why GA4 felt so alien to people who had used Universal Analytics: the model underneath changed, not just the interface.

Engagement time is the newest and the shakiest. Older tools estimated it from the gap between pageviews, which meant the last page of every visit counted as zero. Better tools measure it from real interaction events and stop the clock when the tab is hidden. If your engagement numbers look implausibly high, you are probably measuring how long a tab sat open in the background.

Sources, channels and the direct bucket lie

Where traffic came from is the most decision-useful report in this category and the most misread.

Most tools group referrers into channels: organic search, paid search, referral, social, email and direct. The grouping is a vendor convention rather than a standard, so two tools will assign the same visit to different channels, particularly for anything with a campaign parameter attached.

The important thing to understand is that direct is not a channel, it is a failure bucket. It contains people who typed your address, and also everybody whose referrer was stripped: links from mobile apps, from messaging clients, from PDFs and documents, from email software, and from any site that sets a referrer policy. On many sites the direct bucket is the second largest source and most of it is not direct at all.

A newer report worth having is AI traffic. Visits arriving from ChatGPT, Perplexity and similar tools are a real and growing source, and until recently they landed in referral or direct without being identified. Plausible now breaks this out specifically.

Why no two tools ever agree

Run two analytics tools on the same site for a month and they will disagree, often by a lot. This is normal, it is explainable, and understanding it is more useful than either number.

Consent. If one tool needs a banner and the other does not, the consented tool is missing every visitor who declined. This is usually the largest single cause and it is entirely invisible from inside the tool that is missing them.

Blocking. Ad blockers and browser tracking protection remove some scripts far more aggressively than others. Google Analytics is the most blocked script on the web; a small first-party-served script is blocked much less.

Bot filtering. Around a third of raw arrivals at a typical site are automated: declared crawlers, headless browsers, scrapers from datacentre ranges, link previews nobody saw. Tools differ enormously in how hard they filter, and a tool that filters aggressively will report lower traffic and more accurate traffic at the same time.

Definitions. As above: a cookie-based visitor and a cookieless visitor are not the same unit.

Sampling and thresholding. GA4 applies thresholding to low-volume rows and sampling in explorations, so some numbers are estimates and some rows are simply withheld.

The practical approach is to pick one tool as your system of record, use the others as sanity checks, and never quote two tools' figures in the same sentence.

How it differs from the neighbouring categories

Against product analytics. Product analytics answers questions about identified people over time: of those who did A, how many went on to do B, and which early behaviour predicts whether they are still here in week four. It needs a designed event taxonomy, it costs tens of kilobytes, it sets cookies by design, and it needs an owner. Installed on a marketing site it is heavy, expensive and largely unread.

Against behaviour analytics. Heatmaps and session replay answer what happened on the page rather than how many people arrived. They tell you why a page underperforms, which traffic analytics cannot. They are also the heaviest scripts in this market and, with one exception, they all set cookies.

Most websites need traffic analytics and would benefit from behaviour analytics. Most software products need all three, split across the public site and the authenticated product. Almost nobody needs product analytics on a blog, and a great many people have bought it for one.

What to look for when buying

Five things, roughly in this order.

Does it need a consent banner. This decides more than any feature, because it determines whether your numbers describe your visitors or a self-selected subset of them.

Does it filter bots before counting. A tool that meters raw hits is both less accurate and, on usage-based pricing, more expensive than its headline suggests.

What does it weigh. The spread runs from no script at all to tens of kilobytes, paid by every visitor on every page.

Can you get the data out. CSV export, an API, and a retention period you can live with. A measurement system is supposed to give you a long baseline, so check what happens to your history if you stop paying or change host.

What happens when you exceed the allowance. The published policies differ far more than the products do, and a traffic spike is exactly when you most want to look at your analytics.

The rest of the field

Common questions

Related guides