Download websites that use React

React occupies considerably more conversation than it does web pages. As of October 2026 we detect it on 10,781,865 sites, a smaller footprint than its position in developer discussion would lead you to expect, and understanding why is more useful than the number itself.

Part of the gap is real and part of it is measurement. The real part is that a great deal of the web is not application-shaped: brochure sites, publishers and small commerce do not need a client-side runtime, and the platforms serving them do not ship one. The measurement part is that React is frequently compiled into a bundle that leaves no reliable signature behind, so detection skews toward sites that expose it plainly through recognisable build output or development artefacts.

The honest way to read the count is as a floor. Presence is strong evidence, absence is weak evidence, and any analysis that treats the two symmetrically will be wrong in a predictable direction. We would rather say that plainly than publish a number that implies a census we cannot perform.

What presence does indicate reliably is a build pipeline, and a build pipeline implies people. A site running it was assembled by an in-house team or a technical agency rather than purchased as a template, which is a meaningful qualification for anyone selling developer tooling, hosting, observability, component libraries or engineering services. The population is smaller than the web and considerably more valuable per row.

That same reasoning makes the combinations matter more here than the headline. React alongside a particular hosting provider suggests one kind of team. React alongside a specific analytics account identifies a portfolio of properties run by the same operator. React alongside an unexpected server-side language usually marks a migration in progress, which is a narrow and often lucrative moment to arrive in.

Geographically United States accounts for 9,385,352 detections and the .com zone leads, which follows where engineering teams and venture-funded product work concentrate rather than anything about the library.

The webatla export gives you every domain where we could identify it, with the rest of the detected stack on the same row, the hosting country and the TLD, as one file rebuilt daily. Because the useful queries here are all intersections, holding the whole population locally is the difference between asking three questions and asking thirty.

This measurement humility has a practical upside. Because our detections skew toward sites that expose their build output, the population we do identify is disproportionately made up of properties where an engineering team was working with default tooling rather than an aggressive optimisation pipeline. For most commercial purposes that is the more reachable half of the market, so the bias runs in a convenient direction even though it remains a bias.

It is also worth separating the library from the frameworks built on top of it, because much of what is loosely called React in commercial conversation is really a meta-framework with its own hosting and deployment story. Those are distinguishable in the data by the artefacts they leave behind, and distinguishing them usually matters more than the base library does. That distinction is only available if the export carries every detection on a domain rather than a single primary label.

Meta maintains it under the MIT license and it has been public since 2013. As with every freely distributed library, there is no registry to consult, so a population figure only exists if somebody goes and measures it.