How do AI agents see your website?
More and more of the traffic that decides whether you get bought from is not a person. It is an assistant reading on someone's behalf, or an agent acting with a budget. They do not browse. They fetch a handful of specific files, and if those files are not there, you are not in the answer.
We check the fourteen things they look for, score you out of 100, and hand you the fix for each one — in the order that buys you the most.
What gets checked
Every item here is something an autonomous buyer or a research agent hits in the first few seconds of looking at you. Every one is fixable without redesigning your site.
- robots.txt — is there one, and does it take a position on AI crawlers at all?
- AI user-agents — GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot and peers, named or ignored.
- Content-Signal — machine-readable terms for search, AI grounding and training.
- /llms.txt and /llms-full.txt — the one fetch that lets a model represent you correctly.
- Agent card —
/.well-known/agent-card.json, how an agent learns you can be dealt with directly. - MCP endpoint — can an agent act (book, quote, order), or only read?
- x402 — machine-payable pricing discovery, for buyers that are software.
- Structured data — JSON-LD: who you are, what you sell, at what price.
- Sitemap — present, and actually declared in robots.txt.
- security.txt — the contact a researcher uses instead of a public tweet.
- Soft-404 behaviour — does a missing page say 404, or quietly return 200 and poison a crawl?
- Markdown negotiation — will you serve clean text to a client that asks for it?
- HTTPS & HSTS — transport hardening, redirects, the basics done properly.
- Priority order — what to fix first, weighted by what it is actually worth.
Order the audit
Pay, and the audit runs on its own. No call to book, no questionnaire, nothing waiting on a human being awake.
Payment is handled by Stripe on Stripe's own pages. We never see your card. Price is €99 including any applicable tax, charged once — this is not a subscription.
- You pay. Stripe takes the €99 and sends you back to a status page.
- The audit runs. Automatically, usually within ten minutes. Keep the status page open or bookmark it — the link stays valid.
- You get the report. A score out of 100, every finding with the evidence URL we checked, and a specific fix for each — worst-first.
The report is generated by our own audit engine, not by a person, and it says so. That is the point: it is the same fourteen checks, run the same way, every time — so the score means something and you can re-run it after you fix things and watch it move. If the audit cannot complete — your site is unreachable, or it turns out not to be a public site we can fetch — we refund you in full and tell you why. You are never left holding a €99 charge for a report that did not arrive.
Why we built this
Akshay built this because he kept having the same conversation. A founder with a perfectly good website, real customers, real reviews — and an AI assistant that, asked about them, either invented something or said nothing at all. Not because the business was bad. Because nothing on the site was written for a reader that does not have eyes.
That is a fixable problem, and it is mostly a boring one: a few files, in the right places, saying true things in a machine-readable way. The hard part is knowing which ones and in what order. So we turned the checklist into an engine and put a price on it that does not require a meeting.
Humans: this page. Agents: everything we sell machine-to-machine is at /agents →, and the map of this site is at /llms.txt →.