About Envoyix

How does the audit work?

When you submit a URL, Envoyix fetches four things: the page itself plus /robots.txt, /llms.txt and /sitemap.xml from the same origin. All four go out at once — the three sibling files sit at fixed paths, so there is no reason to wait for the page first. If the page then redirects to a different origin, the siblings are re-fetched from wherever it actually landed, because that is the site an engine would index. The HTML is parsed once into a shared context object, and each check is a pure function of that object. No check performs its own network I/O, which is what makes every one of them unit-testable against a fixed HTML fixture.

Each check returns a score from 0 to 100 and declares a weight. The overall score is the weighted average; category scores are the weighted average within each category. Findings are sorted by impact — weight multiplied by the gap to 100 — so the list you see is ordered by how many overall points each fix would actually recover.

Which crawlers does it check?

Nine, weighted by how much each one affects your chance of being cited. A bot that fetches pages to answer a live question matters more than one building a training corpus, so blocking CCBot costs you far less than blocking OAI-SearchBot.

How does Envoyix identify itself?

Requests are sent with this user agent:

EnvoyixBot/1.0 (+https://envoyix.vercel.app/about; GEO audit tool; respects robots.txt)

Each fetch times out after 10 seconds, follows at most 3 redirects, and reads at most 5 MB of body. If you would rather Envoyix did not fetch your site, disallow EnvoyixBot in your robots.txt.

What happens to the URLs I submit?

Nothing is written to disk and nothing goes into a database. A finished report is held in the server’s memory for 60 seconds, so that a refresh or a second person auditing the same URL is served instantly instead of re-fetching the site; after that it is discarded. There is no account system, no analytics on submitted URLs, and no third party receives them — unless you explicitly enable the optional AI review step on your own deployment, which sends the extracted page text to the Anthropic API using your own API key.

What are the limits of this tool?

• It reads the HTML a server returns to a plain HTTP client. It does not execute JavaScript — deliberately, since that is what most AI crawlers do too — so a client-rendered page scores as an AI crawler would see it.
• Scores are heuristics, not a published ranking formula. No answer engine documents how it selects sources; the weights here encode what is publicly known plus reasonable inference, and they are visible in the source of every check.
• A high score means a page is structurally ready to be cited. It does not guarantee citation, which also depends on the query, the competition and the engine.
• Sites behind authentication, aggressive bot protection, or a geographic block will fail to fetch. That is information too — those barriers apply to AI crawlers as well.

Is Envoyix itself GEO-optimised?

It ought to be, given what it grades. The site server-renders every page, publishes llms.txt and llms-full.txt, ships JSON-LD on every route, and names the AI crawlers explicitly in its robots.txt. Point the analyzer at this page and see.