MACHINE 404 / Custom and headless

Is your custom site readable by AI?

Custom sites are the widest spread we measure. There is no default behaviour to fall back on, so everything present is present because somebody decided it, and everything missing is missing because nobody did.

The single biggest risk: the framework renders everything

A site built as an application often ships an almost empty HTML document and paints the page with JavaScript afterwards. A person sees a finished site. Most AI crawlers see a menu, a footer and an empty div. Open your own page source, not the inspector, and read what is actually delivered before anything runs.

Server rendering or prerendering fixes it, and on a modern framework it is usually a configuration rather than a rewrite. This one check decides more of the score than any other single thing on a custom build.

The soft 404, which only custom sites get this wrong

A catch-all route that answers every unknown address with the homepage and a 200 is a convenience for people and a catastrophe for machines: every URL a crawler guesses becomes a real page in its model of you. It is worth 8 points and it corrupts everything measured after it. Platforms get this right by default; hand-built routers frequently do not.

The advantage nobody uses

A custom site is the only kind where the feed and the agent endpoint are genuinely easy. You already have the data in a database and an API in front of it. Publishing a machine-readable feed is an afternoon, and it is 8 points that a builder site cannot buy at any price. Almost nobody does it, which is exactly why it is worth doing now.