Download websites that use WordPress

About WordPress

Past a certain size a piece of software stops behaving like a product and starts behaving like infrastructure. WordPress crossed that line a long time ago. As of September 2026 we detect it on 23,526,320 sites, and the consequence of that scale is that the interesting economic activity has moved off the platform and onto the layer above it.

There are at least four distinct businesses sitting on that installed base. Hosts sell it as a managed service. Agencies build on it. Plugin and theme vendors sell into it. Security and maintenance firms sell against its failure modes. Each of those has a completely different idea of what a good prospect looks like, and none of them is served by a list of every site running it.

Which is the practical problem with the headline number. A filter on presence alone returns a population large enough to be useless for any purpose except a press release. The queries that actually produce work are narrower and always compound: sites running it alongside a particular ecommerce layer, sites behind a specific proxy, sites in one country on one TLD, sites carrying a plugin with a published advisory, sites that appeared since the previous rebuild.

Every one of those is a filter over a full population rather than a lookup against a directory. The distinction matters commercially, because per-query pricing quietly forbids exploratory work. If each narrowing costs money, you stop narrowing early and settle for a worse list.

The geographic picture follows the same logic. United States holds 9,332,290 of the total and the .com zone carries the largest share, but neither is a targeting instruction on its own. Combined with a platform signal above it and a hosting signal below it, the same column becomes genuinely discriminating.

There is also a time dimension that a single snapshot cannot express. Sites arrive on the platform and leave it, and both directions are commercially interesting: an arrival is a new customer for somebody, a departure is a churned one. Detecting either requires two lists and a diff, which is trivial when you hold the data and impossible when you are querying someone else's.

The webatla export is the full list with the rest of the stack attached to each row, delivered as one file and rebuilt daily. You filter it locally, at whatever depth the question requires, and you can keep yesterday's copy to compare against.

The plugin and theme layer deserves separate mention, because it is where most of the commercially interesting variance lives. Two sites running the same core release can be completely different prospects depending on what is installed on top, and that layer changes far more often than the core does. Detection at that level is what turns a platform census into a segmented market.

There is also a quality caveat worth stating. Sites can hide or rewrite the signals we use to identify the platform and its release, and a small share do exactly that. Our figures therefore describe what is observable rather than what exists, which is the honest position for any measurement of a system nobody is obliged to report to. Where we could not establish a release, the row says so rather than carrying an inferred value.

Our record files the project against WordPress Foundation, it ships under the GPLv2 license, and it has been publicly available since 2003. Like everything else distributed without registration, the only way to know its true footprint is to measure it directly.