Blog

The full rubric: five weighted dimensions, 23 checks and the evidence tier behind each one. Including what the score deliberately leaves out.

A server that only comes back when someone deploys is missing pm2 save or pm2 startup. How to install both, verify each, and test it early.

A stale stylesheet broke our navigation and looked exactly like a failed deploy. Why only a content hash works as a cache-busting token.

We shipped a site in ten languages where nine told search engines not to index them. The bug, the rule that prevents it, and the one-line check.

Every asset came back cf-cache-status: BYPASS while all four cache settings were correct. The cause was a Set-Cookie on every response.

An agent fetches your HTML and throws most of it away. How to hand it clean Markdown instead: negotiation, the Vary trap, .md twins, curl checks.

robots.txt says whether a bot may fetch a page. Content-Signal says what it may then do with it: index, answer, train. Three worked positions.

Complete robots.txt files to copy for the three cases that matter: let everything in, allow search but not training, or block AI entirely.

We measured twelve well-known pages with our own engine. On average 98% of what they send never reaches the model. Here is what survives, and why.

It is usually not the model. Five failure modes in the reading layer, measured on real pages — including the one where rendering the page times out.

Three things decide whether an AI assistant can quote you: robots.txt, JavaScript, and how much content survives extraction. Check all three yourself.

AI agents read your site through a different door than people do. AgentReady measures what they find and tells you what to change first.

Cloudflare's scanner checks which agent protocols your site declares. Ours checks whether your content survives being read. You want both answers.