How website traffic is measured — and why every tool disagrees

The three ways the web gets counted, and how to use any tool's numbers well.

Only one system knows a website's traffic exactly: the site's own analytics, which are private. Every public number — every ranking, every "monthly visits" figure, including ours — is an estimate built from partial evidence. Understanding how the counting works is the difference between using these numbers well and being misled by them.

The three kinds of evidence

First-party measurement. Sites count themselves with analytics tags and server logs. This is ground truth for that one site, but it is almost never public, and even owners argue about it — bot filtering, cookie consent refusals and app-versus-web boundaries all move the total.

Sampled measurement. Public measurement is built from samples: panels of users who agree to be metered, aggregated signals from browsers, extensions, networks and apps, projected from the observed sample to the whole web. Sampling is why public tools can cover millions of sites at all — and why none of them is exact. Small sites are the hardest: a sample that captures a giant reliably may barely see a niche site, so estimates for the long tail carry wide ranges.

Public records. Some facts need no sampling. Domain registries publish when a domain was registered, with whom, and when it expires. Sites publish their own descriptions and pages. App stores publish install ranges, ratings and rankings. These records anchor identity and scale even where traffic itself is sampled.

Why two tools give two different numbers

Put two reputable tools side by side and their figures for the same site routinely differ by 30% or more. The reasons are structural, not sloppiness:

• Different samples. No two panels or signal mixes observe the same slice of the web, and slices skew by country, device and demographic.
• Different definitions. "Visit", "session" and "unique visitor" are defined differently across tools, and some totals fold in app traffic while others count the web only.
• Different bot and spam filtering. A meaningful share of all requests is automated; where each tool draws the line changes the count.
• Modelling choices. Turning observations into a whole-web estimate requires calibration, and honest tools publish a range because the calibration has error bars.

How to use traffic estimates well

Trust comparisons more than points. "Site A is roughly three times site B" survives estimation error; "site A gets exactly 4,213,006 visits" does not — whoever prints it that precisely is rounding a guess.

Trust trends more than snapshots. A site rising for six straight months is real signal even if each month's point figure is approximate, because the same method measures every month.

Compare like with like. Figures from one tool, one definition, one period. Cross-tool comparisons mostly compare the tools.

Mind the shape, not just the size. Sources, countries and device splits explain what a total means. A national champion and a global platform with identical visit counts are different businesses — the giants make this vivid.

What hap.io publishes

hap.io follows the split this page describes. Basic reports publish estimated visits with their range, rank and public-record facts for about two million sites. Full reports publish the measured detail — sources, countries, engagement, trend — only where the measurement supports it, and leave out what cannot be measured responsibly, such as demographics. The about the data page states the same policy in brief, and the glossary defines every figure used.