The score
The score is the share of the pages we opened that are agent-ready, out of the pages we could measure. “5 of 10 pages agent-ready” is 50%. The percent rounds down, so 100% means every page we measured was ready. Pages we couldn’t measure are shown as Not checked and aren’t counted. If we could measure fewer than half the pages, there’s no score: the check says “Not enough pages measured”, and why. A page is agent-ready when it passes both of these:- It opens for every agent we try: a desktop browser, and each of the three assistants that fetch a page for someone.
- None of the four browser checks found a problem.
What we check
The last four are measured with Lighthouse in a desktop browser, because browser agents mostly drive desktop windows. We count the elements that fail, so the check’s tile can say “14 unnamed controls on 4 pages”.
llms.txt is listed under What we checked as one line: present, or not. We check that
/llms.txt is a real text file, not your site’s HTML served at that address. It’s shown for information. It doesn’t change the score and is never a to-do.
Browser access
Browser access on the check’s page shows how many of the sampled pages opened for each agent.
Hover a result to see what got in the way, such as “HTTP 403 on every page”. If an assistant is refused on every page, you get one site-level to-do for it, instead of the same finding on every page.
A page can only be agent-ready if it opened for every agent. So “Perplexity-User: 0 of 10 open” can only sit beside a 0% score.
We make these requests from our own servers. Bot protection judges more than the user agent, so a challenge page our request meets isn’t always a wall a real browser meets. For the browser, we only count a wall when a real Chrome (the one Lighthouse uses) couldn’t load the page either. The three assistants are verified by their companies’ network addresses, so when our copy of one is challenged we mark the page Not checked instead of blocked. A refusal that targets the assistant’s user agent, such as a 403, is counted.
Which pages we open
Agent Readiness opens a sample of your pages, sized by your plan: 3 on Free, 10 on Starter and 50 on Agency. The score is the share of that sample, not of your whole site. The sample is filled in this order:- Your homepage.
- The pages AI cited most in the last 30 days.
- The rest of your sitemap, in order.
What we can’t see
A check that doesn’t click can’t see menus that only open on hover, steps behind a sign-in, behavioral CAPTCHAs that appear after a few clicks, or limits per visitor. The check’s page says so once, and doesn’t score what it can’t measure.Fix a to-do
Select a row in To do to expand it. It shows what we found, what we measured and the pages affected. The steps start on Starter. Agent Readiness to-dos don’t have a snippet or Recheck. Start another check after you make the change.Scope and allowances
Agent Readiness has its own monthly page allowance, the same size as Site Readiness’s, and it only uses the pages it opens. A check never costs more than your plan’s sample.
A scheduled check opens your plan’s sample, and doesn’t use the scope you set. See Limits and allowances.