ChatGPT, Claude, Perplexity and coding agents already visit your site on your customers' behalf. Enter its address: in under 30 seconds you know what they find there and what to fix.
Your website's, for example yourcompany.com. Only the root of the site is tested.
16 checks on public pages: home page, robots.txt, sitemap, llms.txt and the standard files of agent protocols.
A score out of 100, a level from 0 to 5 per category, a plain-language explanation and instructions ready to paste into your coding agent.
Four scored categories, and a fifth for information. The heaviest checks are the ones that matter for every website: being found, being read, letting crawlers in.
Can agents find your pages?
robots.txt is your website's house rules for bots. Search engines and AI assistants read it first to learn where they may go.
A sitemap lists all your pages. An agent finds in a single request what it would otherwise discover link by link, and sees what changed recently.
llms.txt is a summary of your website written for AI models: who you are, your key pages and their links. It is the simplest way in for an assistant.
An HTTP Link header on the home page tells agents, before they even read the page, where to find your useful resources: a summary for AI, a Markdown version, your API description.
Can they read your content without a browser?
Many agents do not run JavaScript. If your home page only shows its text once a script has run, they see an empty page.
Agents prefer Markdown to HTML: less noise, lower cost. When they ask for it, your site can serve them the same page as clean text.
Do you let them in, and on what terms?
Beyond the rules, we check that an AI crawler visiting your home page actually gets it, with no block and no "are you human?" challenge page.
Does your robots.txt allow the main AI crawlers: GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot and Google-Extended? A blocked crawler does not cite your site in its answers.
The Content-Signal directive, placed in robots.txt, states how your content may be used: search, AI assistant answers, model training.
Do you offer services agents know how to use?
MCP is the protocol that lets AI assistants use tools. A card published on your site tells them an MCP server exists and how to connect to it. Useful if you offer a service or data.
A2A lets agents talk to each other. An agent card describes what your agent can do and how to reach it.
Skills are reusable instructions for agents. An index published on your site lets agents discover the ones you offer.
If your service has user accounts, an agent needs to know how to get authorization. The OAuth or OpenID discovery document gives it the right addresses.
This document tells an agent which authorization server protects your API, so it knows where to ask for access.
An API catalog lists your public APIs and their documentation at a standard address, so an agent finds them without searching.
Can an agent pay you? For information, not scored.
The x402 protocol lets an agent pay for content or a service directly over HTTP. It is still emerging: we flag it for information, with no effect on the score.
It is a website that AI assistants and agents can find, read and understand with no effort: a clear robots.txt, a sitemap, an llms.txt summary, text present in the HTML and open access for AI crawlers. For online services it also covers protocols like MCP, so an agent can take action.
Your customers increasingly ask ChatGPT, Claude or Perplexity to compare, summarize or book for them. If these assistants cannot read your website, they cite your competitors. The basic fixes often take less than a day.
The test only reads public pages, as a visitor would, with no login and no form. The report stays in memory for a few minutes so the same test is not run twice, then it is forgotten. No account, no email: we only measure the tested domain name and the score, to improve the tool.
Each check has a weight, higher for the essentials: readable content, AI crawler access, robots.txt and sitemap. A category gets the share of points earned, and a level from 0 to 5. The overall score combines the four scored categories. Checks that do not apply and x402 commerce do not count.
Every point to fix comes with a simple explanation and the fix to apply. The copy button prepares instructions to paste into a coding agent like Claude Code, Cursor or Codex. And if you would rather have help, you can book 30 minutes with us from the report.