An Agent Readiness Score answers one question on a 0–100 scale: how legible and operable is this page to a machine acting on a user's behalf? One number is useful for tracking and comparing. The seven categories underneath are what tell you what to fix.
1. Structured data
Can an agent extract the facts — product, price, availability, organization — from schema.org markup instead of guessing from prose? Missing or malformed JSON-LD is the single most common high-severity failure we see, and for commerce it's decisive: an agent that can't read your price can't buy.
2. Machine-readable forms
Forms are where tasks complete or die. Fields need real names, labels, and autocomplete tokens; a placeholder is not a label. This category punishes the exact failures that silently kill agent signups and checkouts.
3. Semantics & rendering
Real buttons, real landmarks, content present in the served markup. If your critical content only exists after client-side rendering, a large class of agents never sees it.
4. Agent directives
llms.txt, robots.txt posture toward AI crawlers, machine-readable hints about what your site can do. This is where you either invite the traffic or leave it to chance.
5. Navigation
Agents follow hrefs. Navigation that only works via JavaScript click handlers is a dead end, and a site whose pages can't be discovered can't be operated.
6. Metadata & context
Titles, descriptions, canonicals, Open Graph, language attributes — the cheap, boring context that tells a machine what a page is before it commits to reading it.
7. Content legibility
Alt text, descriptive link text, coherent heading structure. Individually low-severity; collectively the difference between a page a machine skims accurately and one it misreads.
How to read your number
The score is a diagnosis, not a grade. The category breakdown is the treatment plan.
Two sites can both score 70 with completely different problems — one needs schema, the other needs form labels. And because the scoring is deterministic and versioned, the number is stable enough to do real work: track it over time, compare against competitors, and gate deploys so it can't silently regress. A score you can't reproduce is a vibe; a score you can gate on is infrastructure.