Methodology

This page explains how a check is performed, so that a score can be reproduced rather than taken on faith. If a number here disagrees with what you see on your own site, the method below is what we will look at first.

What happens when you submit a domain

1. Fetch. The homepage is requested over HTTPS, following redirects, with a bounded timeout and a normal browser user agent. The final URL is the one scored.
2. Read the response. The raw HTML of that first response is kept. Everything in group A and B is decided from it, without executing any JavaScript.
3. Render only if needed. If the first response looks like an empty shell — a client side application with no text in the HTML — the page is rendered once in a headless browser and the rendered HTML is compared with the raw HTML. That comparison is the A2 verdict. It is the only step that costs money, which is why it is conditional.
4. Analyse. Headings, metadata, JSON-LD, link targets, dates and the placement of the opening conclusion are extracted from the HTML that was actually served.
5. Score. Each of the fifteen checks returns pass or fail with the evidence that decided it. The report keeps that evidence, so "failed A4" always comes with the specific link that did not resolve.

What we deliberately do not do

• We do not log in, submit forms, or click. Only what an anonymous crawler can reach is measured, because that is what the AI crawlers see.
• We do not guess. A check we cannot evaluate — a timeout, a TLS handshake we cannot complete — is recorded as unreadable, not as a failure. An unreadable check is our problem, and it is excluded from nothing: it is shown as unreadable in the report.
• We do not score a page we could not fetch. No score is invented from a partial fetch.

One page, or the whole site?

By default a check measures the homepage, which is what an AI engine reads first and what it most often quotes. A site whose homepage is clean and whose product pages are not is not a clean site, and the report says which URL was checked so the scope is never implied.

Repeat checks and comparability

A recheck runs the same fifteen checks against the same URL. Score history is only meaningful when the definition is fixed, so the standard version is recorded with every check and a change in the standard is visible as a change in the standard, not as a drop in your score.

AI assistance, and where it is allowed

Three checks in group C — the ones about whether a passage is genuinely quotable — can use a language model as a second opinion. Where that happens the item is marked as AI-assisted in the report, together with the confidence, so a judgement call is never presented as a measurement. The reachability checks (group A) are mechanical and are never decided by a model: if the crawler is disallowed, no opinion changes that.

What the score is not

It is not a ranking, not a prediction of traffic, and not a judgement about the quality of your business. It measures one thing: how readable this page is to a machine that answers questions.