Website readability for AI agents

We request each page once, with JavaScript off, and read robots.txt, sitemap.xml and llms.txt alongside it. For every domain we take the homepage plus up to six pages a crawler could actually find from there, and blend them into a single score across five pillars.

None of this is a model’s opinion. The rubric is arithmetic, so the same site on the same day always comes out the same, and each point traces back to a check you can see in the report. Blocking training crawlers costs nothing here. That’s a rights decision, not a readability problem.

What this measures is whether a machine can read you, not whether it trusts you. Reputation drives most real citations, and a single fetch can’t see any of it. A high score is the floor, not a promise that anyone quotes you.

Both lists are judgement calls, and they reflect where things stood on July 27, 2026. Sites change. Open any row and re-run the report yourself if you want to check our work.