The first thing we check, because it fails most often
We request your site while identifying as each engine in turn, using their real identifiers: GPTBot, ClaudeBot, PerplexityBot, Googlebot. Then we look at what code each one receives.
Many hosting providers block AI crawlers by default, without telling anyone. The engine gets an error, never reads a single word, and your site simply does not exist for it. This is not a penalty and not a content problem: it is a closed door. And you cannot see it from your browser, because for you the site opens perfectly.
And then, eight more checks
- The raw HTML, with no JavaScript executed: that is how you are read. If the text only appears once something is clicked, they never see it.
- The languages: each with its own address and its hreflang links, or only one gets indexed.
- Structured data: what you tell the machine you are, in its own language.
- The canonical address: pointing at the domain that actually serves the site, not the one that redirects.
- Pages that do not exist: if every one answers "fine", the engine cannot tell where your site ends.
- Weight and speed of what must be downloaded before a single sentence can be read.
- Citable content: direct answers and frequently asked questions, which is the shape these engines quote.
- Your reputation off your own site: what is said about your brand where you are not the one writing, and which namesakes you get confused with.
What you get, and what happens next
A scored report with evidence — the exact command we ran and the exact response we received, so nothing rests on taking our word for it. A plan ordered by impact, not by difficulty. And the files already written: robots.txt, sitemap, structured data.
If we run the site, we apply them. If someone else does, they can apply them as delivered, with nothing to translate.
And it does not end at delivery: these settings change on their own — a provider updates a policy, an engine changes its identifier. That is why the review is periodic: each month we run it with a tool of our own and read it ourselves, telling you immediately if something fails, and each quarter you get in writing whether anything changed — we don’t write just to write.