1. We read your site like an agent
We read robots.txt, your sitemap and up to 25 public pages, choosing the ones closest to buying, booking, joining or getting in touch. On every page we record the forms, buttons, widgets, structured data and any bot wall.
2. We map your journeys
A journey is a way a customer converts: buy, join, sign up, start a trial, book, request a demo, get a quote or get in touch. A journey we opened is found; one we only saw a clear link to is found, not tested.
3. An AI agent walks them
When you ask, an AI agent (Claude, by Anthropic) tries the two journeys that matter most, in a real browser, as a customer’s assistant would. It may click and read, but never type, submit, sign in or buy. It stops at the step where the customer would enter their own details, pay or confirm.
Then we check its claim against what the browser actually saw: a form waiting for the customer’s details, a payment field, or the final button it refused to press. If we cannot confirm it, the result says not verified.
4. The verdict
- Agent ready: every journey we walked reached its final step, verified, and nothing fails in getting in, journeys or forms.
- Partly ready: some walks finished and some did not, or a serious problem will trip agents up.
- Not ready: no journey an agent could follow, or every walk failed.
- Blocked: a bot wall or robots.txt stopped us before any journey.
- Looks promising / Needs work: journeys found but not walked yet, judged on what we could see.
We do not turn this into a 0–100 score. A journey either works for an agent or it doesn’t, and the report shows why.
How accurate is it?
We measure it. On 18 November 2026 an AI agent walked 23 hand-checked journeys twice each: 19 on 17 public websites in 8 industries (ecommerce, fitness, healthcare, services, finance, software, nonprofit, education) and 4 on our own test site. All 98 checks were correct, and both runs agreed on every journey.
We also check what the crawler finds against 45 hand-checked websites. All 500 checks we could observe were correct. 41 more could not be observed and are not counted.
Anything we could not test is marked inconclusive, never passed.