Download websites that use Tailwind CSS

About Tailwind CSS

Almost everything we detect on a page is something the page loaded. Tailwind CSS is different, because in its normal configuration it is not loaded at all. The framework runs at build time, scans the source for class names, emits a stylesheet containing only what was used, and then disappears from the delivered artefact except for the fingerprint it leaves behind in the markup. Detection therefore rests on inference rather than on a file we can point at, and the honest way to describe our count is as a lower bound.

As of September 2026 we see it on 1,028,347 sites, filed under UI frameworks. The evidence is usually the shape of the class attributes, the naming and ordering conventions the compiler produces, or a stylesheet whose contents match generated output. Sites that rename classes, run an aggressive post processor, or ship a heavily customised token set can pass through undetected. Nothing in the pipeline recovers those, so the figure describes sites where the fingerprint survived, not sites that use the framework. Anyone sizing a market from this column should treat the gap as real and probably concentrated among the larger and better resourced sites, since those are the ones most likely to run a build step that erases the evidence.

Versions are the part of this record people misread. We resolve 124 distinct version strings across the population, and 2.2.19 is a value that recurs often. A version string here describes the toolchain that last built the site rather than anything the site is running today, because the compiled stylesheet carries the signature of whichever release produced it. A site sitting on an old release is a site whose front end has not been rebuilt in a long time, which is frequently more commercially interesting than the framework itself. Agencies hunting rebuild work, and vendors selling anything that requires a redeploy, are effectively buying a staleness index.

United States carries 610,356 of the detections and the extension mix concentrates on .com, which together describe a developer facing, commercially hosted population rather than a national one. Tailwind Labs released the framework in 2017 and publishes it under the MIT licence, so there is no purchase event to look for and no licence server to correlate against. Adoption tells you about the taste and the recency of a development team, not about a budget line, and the two are worth different amounts depending on what you sell. A component vendor reads the same row as a qualified prospect. A hosting salesperson reads it as noise.

A per domain lookup answers whether one site shows the fingerprint, which is rarely a question anyone is paid to ask. The questions worth paying for are comparative. How much of a given hosting provider's estate has been rebuilt recently. Which sites pair the framework with a content management system that constrains what a redesign can touch. Which agencies leave the same build signature across every client they ship, a pattern that only becomes visible once you can group thousands of domains by a shared fingerprint. Our export carries the domain, the detected version, the rest of the stack and the crawl date, which is exactly what those questions consume. A lookup endpoint cannot group. A file can, and the same file run against last quarter shows you what changed.