Download websites that use AMP

About AMP

A technology normally spreads because it solves a problem for the person installing it. AMP spread because a search engine rewarded it. Publishers rebuilt article templates against a restricted subset of HTML in order to qualify for preferential mobile placement, and when the requirement behind that placement was relaxed, the rebuilt templates stayed where they were.

That inversion is what makes the current population interesting. A site carrying the framework today is maintaining a second rendering path for its content, whether or not anyone still looks at it. The useful question is no longer who adopted the format but who still has it and has not taken it out, because removal requires a template project that returns nothing visible to the business. Detection therefore identifies content management setups that have not been revisited since the incentive existed, and publisher plugins that are still switched on because switching them off was never anyone task.

Origin matters for how the population behaves. The framework came from Google in 2015 and is published under Apache 2.0 terms, which makes it open source with governance that stayed close to a single platform. When the platform priority moved, adoption had no independent constituency to sustain it. Researchers get an unusually clean natural experiment out of that: the decay curve of a standard whose sponsor withdrew the reward, measurable across snapshots rather than argued about.

As of September 2026 we identify AMP on 1,548,329 sites. United States accounts for 1,184,166 of them and the suffix mix leans heavily toward .com, which reflects both the concentration of the online publishing industry and the fact that country in this data means hosting and registration rather than readership.

Detection works from the framework runtime and from the discovery link that a canonical page uses to point at its alternate version. Both signals live in public markup, and both can be missing while the alternate pages still exist. A site that generates the alternate version but stopped linking to it will be undercounted, and a site whose home page never carried the markup can be missed even when every article template does, since a crawl that samples shallowly sees the front door rather than the archive. There is no meaningful version signal to report, so the record is a presence flag and should be read as one.

Checking a single domain answers whether that publisher still ships the format. A list answers the questions people actually pay to have answered. Which publishers, in which countries, are still carrying a legacy rendering path and would take a migration proposal. Which content platforms account for the bulk of the remaining installs, which sizes the plugin base for whoever maintains those platforms. How quickly the population shrinks between one snapshot and the next, which is the only honest way to say whether the format is being retired or quietly kept alive by templates nobody edits. None of those are visible one domain at a time.