See your website the way an AI assistant does.
When someone asks ChatGPT or Claude about your industry, an answer gets written from whatever those models can reach and read. This checks whether that includes you — and writes the files you are missing, from your own words.
Are AI crawlers actually served?
We request your homepage as a browser, then again as GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot, and compare what comes back. A CDN that quietly returns a challenge to crawlers is invisible in every other audit.
robots.txt, read the way a crawler reads it
Group selection, longest-match rules, Allow beating Disallow — and the Cloudflare-managed block that sits above your own file and disallows the AI agents without telling you.
llms.txt and llms-full.txt
Present, well-formed, and actually a file — not your HTML page answering 200 to any path it does not recognise, which is how most sites accidentally 'have' an llms.txt.
Sitemap and freshness
Sitemap indexes, gzipped sitemaps, whether robots.txt declares them, how many URLs carry a lastmod, and whether the newest date makes your site look abandoned.
ai.txt, and what it is actually for
Whether assistants may cite you and whether they may train on you are different questions. We read your ai.txt, flag where it contradicts your robots.txt, and write you one that matches the position you have already taken.
soul.md, memory.md, AGENTS.md
The files people are starting to write for agents rather than for readers. We fetch them, read them, and check nobody has left a credential in one.
Structure, schema and trust
Titles, headings, JSON-LD types, whether anything declares who publishes the site, whether your text survives without JavaScript, and whether a model can find a byline or a date.
Can an AI agent transact with you?
A second, optional audit on the same scan, run only when you ask. We request your pricing page as a browser and again as each agent, check whether the price is text a machine can read, and look for an API that answers a machine properly, an OpenAPI document, an MCP endpoint, a UCP manifest and a way for an agent to obtain its own credentials. A tier, not a score, and it never touches a form or a payment.
The test nobody else runs
Most site audits read your robots.txt and stop. But robots.txt is a request, not a control. Plenty of sites publish a welcoming robots.txt while their CDN returns a 403 or a JavaScript challenge to the same crawlers at the edge — and the owner has no idea, because in a browser everything looks fine.
So we ask twice. Once as a browser, once as each crawler, then compare the status code, the size of the response and whether your own page title survived. If ClaudeBot gets something different from what you get, we show you exactly what and where the difference is.
Then we do something stranger: we hand a language model only the text a crawler would receive, and ask it the seven questions your customers actually ask. What do you do. Who is it for. Where are you. What does it cost. If the model cannot answer from your own pages, neither can the assistant your customer is talking to.
If you own the site, there is a fourth thing to ask for: we go back as an agent rather than an assistant and ask whether it could act. Is the price an agent reads the one you show a person; does your API tell an unauthenticated machine how to behave; is there an OpenAPI document, an MCP endpoint or a UCP manifest to find; could an agent obtain a credential of its own. That comes back as an Agent Commerce Readiness tier, T0 to T5, decided by predicates you can check with curl, and it never moves the score above. We never submit a form or a payment to find out.
What you get
- Free, no accountA Citation Readiness score out of 100, the category breakdown, and your top three problems.
- With a free accountEvery finding with its evidence, a prioritised fix list, the answerability test in full — and, on request, the Agent Commerce Readiness tier for a site you own.
- Your files, writtenAn llms.txt and llms-full.txt built from your own pages — copy them off the screen or download the lot as a zip.
- Beyond the filesStructured data recommended for the content you actually have, and an honest answer on whether an API or MCP server is worth building yet.
Nothing in a generated file is invented. Every line comes from text your site already publishes; where we have no source, you get a visible TODO instead of a guess. How the score works.
Check a site now
About a minute. No account needed for the summary.