About

Take a group of websites (your town's startups, your friends' blogs) and track their traffic relative to each other. That's the whole idea, same as it was in 2008.

Nobody outside a site can measure its traffic. Every published ranking is measuring a proxy: DNS lookups, browser telemetry, backlinks. Each proxy is wrong in its own direction, so we don't pick one. A site's rank here is the plain arithmetic mean of Tranco, Majestic and Umbrella, whichever of them have ranked it. Radar and CrUX sit alongside for context. They report a band, not a place, so they don't get a vote. More on that below.

The 5 sources

Tranco

A consensus list, combining the four below plus Farsight's DNS data. Updated daily.

Radar

What Cloudflare's DNS resolver sees people looking up. Updated weekly, in buckets.

CrUX

Real Chrome users who visited the site. Updated monthly, in buckets.

Majestic

Backlinks: how much of the web points at you, by referring subnet. Updated daily.

Umbrella

Cisco's DNS query volume, which counts machines as much as people. Updated daily.

What the numbers don't say

A — means that source hasn't ranked the site, so we leave it out of the average rather than counting it as a million. The 2008 version of this site did count it as a million, which dragged every average toward a number nobody had measured. A site no source has ranked sorts last, showing —, which is at least honest about being ignorant.

Two of the sources report bands, not places. Radar and CrUX only publish which tier a site fell in. CrUX will tell you a site is in the top 5,000; it will never tell you it is 4,113th. A band cap is not a rank. Average it in at face value and a site truly sitting at 50 gets dragged toward 1,000, the edge of the band CrUX names, which punishes the sites doing best. So Radar and CrUX ride along as their own columns, each showing the band it asserts (top 5k, top 1M), and the mean is left to the 3 sources that count places.

Even those disagree about scale. Majestic ranks the web's opinion of you rather than its attendance. The average of the three gives you a rough number, useful for direction and not much else. The chart is log-scale because ranks in one index can run from 12 to 400,000, and a linear axis would flatten everyone but the leader onto the floor.

The part we should probably admit

Tranco is itself an average. It's a Dowdall combination (a weighted rank vote, not a mean) of CrUX, Majestic, Radar, Umbrella and Farsight. Two of the three sources we average, Majestic and Umbrella, are already folded inside it. So the mean leans on them twice, it's about 1.2x more confident of itself than it looks, and it isn't averaging like-for-like: it's averaging two measurements next to a vote that already counts those same two.

We kept it anyway. Tranco is the only way we get Farsight's data at all (it isn't published anywhere we can reach), and it's the only source with enough history to make the chart real on day one. Also, averaging a number with an average that already contains it is exactly the crime you'd expect from a site whose entire thesis is "all the traffic measures are flawed, so use the average of a bunch of them." We're not going to hide it on the page that exists to explain the flaws.

Editing

Anyone can create an index and add sites to it. There are no accounts. Every change is public in each index's History and one click to undo. Authentication will happen if abuse does.

Some indices mirror a list somebody else maintains, like the official Y Combinator API. Those say so at the top, link the source, and aren't editable here: a sync would clobber your edit on its next run, so the site doesn't pretend to accept it.