How does the check work?

The check works by reading your website the way an AI crawler reads it: the raw HTML your server sends, with none of your JavaScript run, and the three small files at your domain root that tell machines what they may do. It takes about 15 seconds and makes at most 14 requests, which is less than a single visitor loading one page.

What happens when I press Run check?

1. Your address is tidied up. Bare domains, full links, subpages and pasted junk all work. Spaces, tracking codes and stray quotes are stripped, https is tried first, and www or non-www is followed automatically.
2. The page is fetched twice. Once with a normal browser identity, once announcing itself as GPTBot. If the browser gets your page and the AI crawler gets a 403 or a challenge screen, that is a firewall rule aimed at AI, and robots.txt never gets a say. The robots standard itself, RFC 9309, is blunt about the difference.
3. Three root files are read. robots.txt for the crawler rules, the sitemap for how much of a site there is, and llms.txt for the file written specifically for AI systems.
4. 40+ checks run on what came back. Every check records a pass, partial or fail plus a one-line piece of evidence, so the report tells you what was found, not just that something was wrong.
5. The score is written and saved. You get an AIQ out of 100, a letter grade, an identity label and a one-line verdict, then the report link. Two screenshots are taken in the background when the service is available and never hold the report up; a re-check of the same domain within 24 hours reuses them, so a fix shows up in under 60 seconds.

These rules are not a form of access authorization.

What does the checker fetch?

Only public web addresses on the standard ports. The fetcher resolves your hostname before connecting and refuses private or reserved addresses, follows at most 5 redirects while re-checking each hop, gives up after 15 seconds per request, and stops reading a response at 3 MB. It identifies itself honestly with an AgentRank Checker suffix on its user agent, so your server logs show exactly who visited.

How is the score calculated?

Every check carries a weight in points. Your AIQ is the share of the applicable weight you earned, rounded to a whole number out of 100. Checks that cannot apply to your site, for example the mixed-content check on a site that has no HTTPS, drop out of the denominator rather than counting against you. Texas notice checks join the rubric only when a Texas address or license appears on the page. The grade ladder turns the number into a letter and a label, and the full check list shows the weight of each one.

What does the checker never do?

• It never runs your JavaScript, because most AI crawlers do not either. If your content only appears after scripts run, the report says so.
• It never logs in, fills forms or crawls your whole site. One page, three root files, done.
• It never stores a copy of your page. The report keeps scores, evidence lines and two screenshots.
• It never asks for an email to show you the result. See what we store on the privacy page.

Where do the checks come from?

The crawler-access rules follow the published documentation for OpenAI, Anthropic and Perplexity crawlers and the robots.txt standard, RFC 9309. The content-shape checks come from Princeton’s GEO study, which measured which edits moved a page into AI answers. The real estate layer comes from reading a lot of agent websites over 3 years and noticing the same 8 things missing over and over.