Measurements, not model opinions
Every finding is an HTTP fact you can reproduce with curl. We do not ask a language model what it thinks of your brand and chart the answer — we test whether the crawlers behind those models can complete a request and read what comes back.
Fetch as a browser
We request your homepage with an ordinary browser user-agent to establish a baseline: status code, response time, byte size, CDN in front of it, and the visible text in the raw HTML.
Read robots.txt properly
We parse robots.txt following the REP specification — group selection by longest matching user-agent token, longest-match wins between Allow and Disallow — then evaluate the policy for fourteen AI crawlers individually. Retrieval crawlers are scored separately from training crawlers, because blocking them means different things.
Request as the crawlers themselves
We send live requests using the published user-agent strings for OAI-SearchBot, PerplexityBot, ClaudeBot and GPTBot, then compare status codes and response bodies against the baseline. This is what catches CDN and WAF blocking that robots.txt says is not happening.
Render and diff
We load the page in headless Chromium with JavaScript enabled and count the visible words, then compare against the raw HTML. The gap is precisely the content AI crawlers cannot see, because none of them execute JavaScript.
Inspect the machine-readable layer
JSON-LD blocks are parsed and validated, entity types are checked for coverage, and we look at sitemap, llms.txt, canonical, headings, meta and question-shaped content.
Score and explain
Findings are weighted across access, readability, structure and answerability. Every finding carries the evidence behind it, so you can reproduce it yourself.
What we do not claim
We do not measure what ChatGPT says about you, how often you are mentioned, or your share of voice against competitors. Those require querying the assistants themselves, and any tool doing it is sampling a non-deterministic system.
We measure the prerequisite instead: whether the crawlers can reach you and read you. That layer is deterministic, and if it fails, nothing downstream is fixable.
Try it on your site
No signup. No credit card. Results are a shareable link.