For sites that say they are AI-native
Can an AI actually read your site?
Agents and crawlers get the HTML your server sends and nothing else. This reads a page the same way and reports what it found, check by check. No score, because the one that fails is the one that matters.
What a crawler receives
What is published for machines
If the served words are thin or the canonical points elsewhere, the fix is on the server, not in the prompt. If the words are there, give the agent a DESIGN.md next, so what it builds follows a system instead of guessing one.
Why this exists
While researching agencies that call themselves AI-native, we found one whose portfolio, product and contact
pages each carried a canonical pointing at the homepage, whose every URL answered 200, and whose
value proposition appeared zero times in the served HTML. Nobody running it had reason to suspect any of
that. The first version of this very check nearly made the same mistake in reverse: it saw a 200 on
/llms.txt and believed it, until a request for a path that cannot exist also returned 200.
Every presence check here carries that control.
Questions
What does this actually test?
The HTML your server sends, exactly as a crawler or a coding agent receives it, with no JavaScript run. It counts the readable words, checks the title, H1, meta description and canonical, reads any JSON-LD, then probes robots.txt, llms.txt and ai.txt. Every presence check is judged against a control request for a path that cannot exist, because a site that answers 200 to everything answers 200 to /llms.txt too.
Why is there no score?
Because a single number hides the one check that fails, and the failing check is the useful part. A site can pass eleven checks and still tell search engines its product page is a duplicate of its homepage. You would never see that behind a 92.
What can it not see?
It does not render the page, so it cannot measure how many words appear only after JavaScript runs. It reports the served word count and flags a client-side shell when a document is almost all script, which is the honest proxy. It also reads only the URL you give it, not the whole site.
Is anything stored?
No. The check runs on request, the result is returned to your browser and nothing about the URL or the result is written anywhere. Requests are rate-limited per address so the tool cannot be used to hammer a third party.