Your site now gets an agent-readiness score. Here is what it measures.

What the standard asks for

The AgentReady spec was published in August by Ora and Vercel, with Mintlify and Auth0. It sorts requirements into three stages, and it is refreshingly blunt that it is a list of requirements, not a score.

That last row is most of the difference between a good score and a bad one, and a language school or a manufacturer will never tick it. Read the grade with that in mind.

The part worth trusting

The standard ships the data it was built on: 1,033 agent sessions across 25 sites and 190 controlled fetch tests. We recomputed the headline numbers from those files rather than taking the launch post at its word.

Two findings held up, and they are the two that matter.

Blocking an agent is the only failure with no exceptions. In the fetch tests, a blocked page failed every single time, for every kind of client. Nothing else in the standard comes close to that.

A page that draws its answer with JavaScript is a gamble. The same page failed every time for a simple fetcher and passed every time for an agent that runs JavaScript. So it does not break everything, it breaks the cheap half. If your product pages need JavaScript to show what you sell, you are betting on which agent turns up.

The part that does not transfer

The headline doing the rounds is that documentation pages produced 47% of the answers agents gave. That is true of this study, and it tells you almost nothing about a brand site.

The 25 sites are all software companies: Stripe, Twilio, Notion, Snowflake, Datadog and the like. The tasks were developer tasks. Documentation won because documentation was the thing worth reading. If you sell language training or industrial parts, your equivalent is a commercial page that answers a buying question, and no scanner will tell you whether you have one.

What it says about llms.txt

The standard lists llms.txt as a SHOULD, which sits awkwardly next to what we published last week. Its own data settles it more calmly than either side would like. On sites that ship the file, agents fetched it in 39% of sessions, and it was the source of the final answer in about 12% of sessions overall.

Those agents were reading developer documentation to complete a task. They are not the answer engines your buyers use, which still do not fetch the file. Both things are true: worth having if you publish technical docs, not a visibility lever for a brand.

What I would do with a score

Run it, because it is free and it takes a minute. Then treat it as a checklist rather than a result.

Fix the two hard failures first: anything in robots.txt blocking an agent you want, and any commercial page whose content only appears after JavaScript runs. Add a sitemap and clean structured data if they are missing. Ignore the API and MCP items unless you sell a developer product, and do not let the score be reported as AI visibility, because it measures whether a bot can use your site, not whether an engine recommends you.

For what it is worth, this site scores 88 out of 100 on Ora's scanner. That took no special work. The pages are server-rendered and nothing is blocked.

The bottom line

The scanners are a useful free floor, and the standard is honest about being a list of requirements. Clearing the floor is not the same as being recommended, and only one of those shows up in your pipeline.