Download websites that use Bootstrap

About Bootstrap

Bootstrap solved a problem that has largely stopped existing. Consistent layout across browsers, a grid that behaved predictably, and components that looked deliberate without a designer were genuinely hard when it appeared in 2011. Most of that is now native to the platform.

And yet, as of September 2026, we detect it on 14,017,821 sites. Layout code that ships inside a theme outlives the fashion that produced it, because removing it requires a rebuild and nothing forces a rebuild.

That persistence is what makes it useful, and the use is not really about the library. A site carrying it is usually a site that was built once, from a template or a starter theme, by someone who has not returned to it since. That is an inference about maintenance behaviour rather than about technology preference, and maintenance behaviour is what a large share of commercial research is actually trying to establish.

The inference gets much stronger in combination. On its own, presence is weak evidence: plenty of actively maintained sites still carry it deliberately. Alongside an old interpreter, a dated analytics generation and a CMS release from several cycles back, it stops being ambiguous. Each signal is individually noisy and the stack as a whole is not, which is the general argument for holding all the detections on one row rather than querying them one at a time.

This is also why single-technology directories tend to disappoint. They can tell you a site uses something. They cannot tell you what else it uses, which is where nearly all of the discriminating power lives.

The geographic picture is consistent with the age reading. United States carries 5,416,257 of detections and the .com zone leads, both tracking the markets where template-built commercial sites accumulated fastest.

For agencies, the practical output of all this is a redesign prospect list: properties that are commercially active enough to run analytics, old enough to show a dated front-end stack, and located somewhere you can sell into. That is three filters over one population, and it is cheap only if you hold the population.

The webatla export delivers exactly that. Every domain we detect it on, with the rest of the identified stack on the same row, the hosting country and the TLD, as a single file rebuilt daily and filtered on your own machine.

The counter-argument deserves airing, because it limits how far this inference can be pushed. The library is still chosen deliberately by teams who value predictability over novelty, and internal tools and documentation sites use it precisely because nobody wants to design them. So a filter on presence alone will pull in a meaningful minority of well-maintained properties. That is why the maintenance reading only holds up in combination, and why we publish the whole stack per row rather than the single detection.

The version data has a similar caveat here. Layout libraries are frequently vendored, renamed or compiled into a larger bundle, so the release we observe is less reliable than for software that announces itself in a header. We report what we saw rather than normalising it into something tidier, because a tidy number that quietly guesses is worse than an untidy one that does not.

It is distributed under the MIT license and has been public since 2011, with no registration step anywhere in its history, which is why measurement is the only route to a population figure.