# About this tool

AI systems like ChatGPT, Claude, Gemini, and Perplexity increasingly answer
questions about your site. canaisee runs a battery of checks to show how
readable your site is to them, and what you can do to improve that readability.

## What we measure

Checks span four dimensions: crawler accessibility, content readability,
semantic structure, and agent-native surface. v1.3 groups them into two
tiers — `Hygiene` (table stakes) and `Frontier` (ahead-of-the-curve
agent-native features). Every check is open — you can see the evidence and
the flags behind each score on the scorecard itself.

## What we can't measure

canaisee cannot directly verify whether a specific model has ingested your
content, whether ChatGPT cites your site in answers, or whether your pages
appear in any given training corpus. We measure signals that should
correlate with those outcomes. For outcome-level questions see the
companion "What does AI say about your site?" demo on the Evangent site.

## This is an opinionated beta, not a settled standard

The grade is a transparent, opinionated read on a fast-moving surface.
Some of what we measure — `robots.txt`, structured data, heading
hierarchy, sitemaps — sits on stable specs with broad adoption. Some —
`Accept: text/markdown`, `/.well-known/mcp.json`, WebMCP,
`Content-Signal` — is actively emerging and not yet universal. We include
the latter because agent-first publishers are adopting them and they're
cheap to add. The full [rubric](https://canaisee.com/rubric) is published
so the opinion is inspectable, argue-able, and versioned as the web changes.

## Built by Evangent

canaisee is built by [Evangent](https://evangent.org). It's free, runs
without signup, and the rubric is published and versioned so anyone can
see how the grading works. Source code lives at
https://github.com/8gara8/agentread under the MIT license.