- Home
- Methodology
Methodology
Where the numbers come from, how they are measured, and what they cannot tell you.
An SEO tool that does not say where its data comes from is asking you to take its word for it. This page sets out the sources, the methods and — the part usually left out — the limits.
Three kinds of source, and only three
Direct measurement. The tool connects to the target itself and reports what it observes: HTTP headers, TLS certificates, page weight, redirect chains, robots.txt. These results are exact and reproducible. Most tools on this site work this way.
Official APIs. Google PageSpeed Insights and Chrome UX Report for performance, Google Safe Browsing for malware, public DNS resolvers, and registry RDAP services for domain records. The provider is always named on the result page.
Local computation. Word counts, hashes, encoding, colour contrast, file generation. Nothing leaves the server, nothing is fetched, and the result depends only on what you typed.
There is no fourth category. No proprietary index, no invented authority score, no estimate dressed up as a measurement.
What the tools cannot do
Rank tracking is a snapshot, not the truth. Google personalises results by location, device, language and history. A rank check runs from a neutral session with no history, which is the most reproducible baseline available — and still not what any specific person sees.
Backlink data is always partial. No one has a complete map of the web's links, including the largest commercial providers. Figures reported here come from a named provider and represent what that provider has crawled — not what exists.
Domain authority scores are third-party inventions. Google publishes no such metric. Every authority score in existence is a vendor's model, useful for comparison within that vendor's scale and meaningless outside it. This is stated on the tool rather than implied.
Field performance data has a lag. Chrome UX Report figures are a rolling 28-day average at the 75th percentile. A fix deployed today will not show up for weeks. Lab measurements react immediately but reflect one synthetic run, not your visitors.
Screenshots are unavailable. Rendering a page as an image requires a headless browser, which shared hosting cannot run. Rather than substituting something else and calling it a screenshot, the tool says what it cannot do and reports what it can measure without rendering.
How the tools behave towards other sites
Every request identifies itself with a clear user agent. Crawling tools honour robots.txt. Requests to the same host are throttled to one every 250 milliseconds. Redirects are followed one at a time and revalidated, and requests to private network addresses are refused — a tool that fetches arbitrary URLs must not become a way to probe internal networks.
Result pages are never indexed
Every result page carries noindex, both as a meta tag and as an HTTP header.
With 53 tools, indexing results would generate thousands of thin, near-identical
pages — precisely the pattern that damaged this domain under its previous software. Results
are for you, not for search engines.
Rate limits
Tools that rely on a paid data provider are limited to three uses per day per IP address. Everything else allows 300, which exists only to stop automated abuse. IP addresses are hashed with SHA-256 before being written, so the counter works without the site holding identifiable data.
This English section
12 of the 53 tools are currently available in English. Each page is written, not machine-translated: the tools' explanations, warnings and result text are authored in English rather than passed through a translation engine. This is partly a quality judgement and partly a compliance one — Google's spam policies list text translated automatically without human review as a violation.
New tools are added to the English section based on what people ask for. The complete catalogue exists in French.
Corrections
If a result looks wrong, it may well be. Report it through the contact page with the address you tested and what you expected. Measurement errors are fixed as a priority, and a tool whose source disappears is removed rather than left running on stale assumptions.