"How do I check website traffic" is two questions wearing one coat. Your own traffic can be counted exactly. Everyone else's can only be estimated, and those estimates are more useful than people expect for large sites and much less useful for small ones.
A split diagram. Checking your own website traffic gives an exact count from a script on the page. Checking someone else's gives a modelled estimate built from panel data, with an error range that grows as the site gets smaller.
Your own site: just count it
Put a script on your pages. It fires when someone loads one, the page gets counted, and what you read in the dashboard is a count rather than an inference. Ours is 1270 bytes and takes about two minutes to install, but any analytics tool works the same way at this level.
There are two other sources worth knowing about, and both are partial.
Your server logs record every request that reaches your server. That sounds complete and isn't: requests aren't people, a single page view is a dozen requests, bots make a huge share of them, and anything served from a CDN cache may never appear at all. Logs are useful for debugging and poor for counting.
Google Search Console shows impressions and clicks from Google search. That's a genuinely accurate slice, but it's a slice — nothing from direct visits, social, email, or any other search engine. People regularly mistake it for their whole traffic and conclude their site is a tenth the size it is.
A comparison of three sources for your own traffic. An analytics script counts every pageview from every source. Server logs count requests including bots and miss cached responses. Search Console covers only visits arriving from Google search.
Someone else's: how the estimate is built
Nobody outside a website can see its real numbers. There's no public register and no back door. So every tool offering you a competitor's traffic is modelling it, and they all do it roughly the same way.
They recruit a panel — people who've installed a browser extension, or whose data comes bundled from an app or an ISP — and they watch where that panel goes. Then they scale it up. If 40 people out of a two-million-person panel visited a site last month, and the panel is assumed to represent one in every few thousand internet users, you multiply out and publish the result.
A diagram of how traffic estimation works. A small panel of tracked users is observed visiting a site, then that count is multiplied up to represent the whole internet. A large site is visited by many panellists so the estimate is stable; a small site is visited by only one or two, so the multiplier amplifies a tiny sample into a wildly uncertain figure.
The figures in that drawing are illustrative, but the shape of the problem is real and it's the thing to take away. Estimation error isn't a flat percentage applied evenly. It's a function of how many panellists happened to visit, so it's small for big sites and enormous for small ones. A site doing a few thousand visits a month might have zero panellists in a given period, at which point the tool either reports nothing or reports something invented by a model filling in a gap.
This also explains a thing people find baffling: estimators are often wildly wrong about the site you know best, which is usually your own or a friend's — because those are small. The tool looks broken. It isn't. You're just standing in the part of its range where it doesn't work.
What estimates are good for
This isn't an argument that they're useless. Used for the right job they're the only option there is.
- Relative size between large sites. Is this competitor roughly ten times us or roughly the same? That survives the error bars.
- Direction of travel. A site estimated as climbing steadily for a year is probably climbing, even if the absolute number is off, because the bias is fairly consistent month to month.
- Channel mix. Which share is search versus social versus direct is often more reliable than the total, and more useful for working out what somebody is actually doing.
What they're not good for is a number in a board deck without a range attached, or any conclusion about a site doing under roughly fifty thousand visits a month.
If it is your own site
Then skip all of the above, because you can simply count. One script tag, and the numbers are counted rather than modelled — every pageview, every source, no sampling and no gap-filling. We never store an IP address and set no cookies, so there's no banner standing between you and the data either, which matters more than it sounds: a banner costs you a share of the very visitors you're trying to count.
If you're moving from somewhere else, what imports and what can't is listed in full before you start. If you're still deciding, there's a page about how to judge one. And if the number you're worried about is bounce rate, that one has its own problems.
The short version
- Your own traffic can be counted exactly. Everyone else's can only be modelled.
- Estimator accuracy tracks site size. Big sites, usable. Small sites, not.
- Search Console is a slice of your traffic, not all of it. Server logs count requests, not people.
- Use estimates for relative size, direction and channel mix. Not for a figure you'd defend in a meeting.