There's a lot of hand-waving about "AI-ready" websites — most of it selling a plugin or a text file that does nothing. But Google has quietly published something concrete: a spec for building websites that AI agents can actually use, on web.dev. And unlike almost all SEO advice, it's a checklist with a genuine pass/fail. So rather than tell you my own site is agent-ready, I ran the checklist against it and I'm showing you the results — including the parts that failed the first time.
Why this checklist is different from the usual advice
The thing an AI agent "reads" when it looks at a web page isn't the pretty layout — it's the accessibility tree: the underlying structure of roles, names and states that screen readers have used for twenty years. Which leads to a genuinely useful conclusion: building a site for AI agents and building one for people using assistive technology are now the same engineering job.
That matters for two reasons. First, it's testable — most SEO advice is vibes, but this is a pass/fail you can check. Second, it's not a gimmick that dates: a study from Berkeley and Michigan found agent task-success dropped from 78% to 42% on sites with a degraded accessibility tree — nearly half of all failures caused purely by missing structure. Get the structure right and you serve agents, screen-reader users and search engines with one piece of work.
The checklist — and how my site scored
Here's Google's spec, item by item, against rgleeson.com:
- Content in the raw HTML, before any JavaScript runs. ✅ The homepage carries 735 words of real content in the initial HTML; every page is static. This is the big one — most sites that fail AI crawlers fail here, so it's the first thing I check.
- Semantic HTML — real
<button>and<a>, not styled<div>s. ✅ Every interactive element is a proper button or link; zero "fake" clickable divs across the site. - Interactive elements large enough to see and click. ✅ Controls meet the 24px target. (This one failed first time — see below.)
- Form fields with associated
<label>s. ✅ Every input is linked to its label, so an agent can tell what each field is for. - A consistent, stable layout across pages. ✅ The navigation and footer sit in the same place on every page, so an agent reading screenshots isn't thrown by things moving around.
- No "ghost" elements or transparent overlays hiding controls. ✅ None — visual analysis discards anything it thinks is covered, so there's nothing floating invisibly over the buttons.
- A clear signal that a thing is clickable. ✅ All 37 interactive elements on the homepage use a pointer cursor — a documented actionability cue.
- Landmarks and a skip link. ✅ Every page has a
<main>region and a "skip to content" link. (This one also failed first time.) - A visible focus state for keyboard and agent navigation. ✅ Now present site-wide. (Failed first time too.)
The bits that failed — and why I'm telling you
It did not all pass on the first run, and that's rather the point.
The testimonial carousel's arrow buttons had quietly collapsed to two pixels wide — a subtle interaction between an inline icon and the layout meant they were rendering as slivers. To a person they were fiddly; to an agent's visual analysis, below the minimum size and effectively invisible. Several pages were also missing a <main> landmark and a skip link, and there was no visible focus outline anywhere. All fixed now — but I only knew because I measured.
That's the real lesson. This is a hand-built static site by someone who does this for a living, and it still had four things to fix. "Probably fine" isn't a claim you can stand behind. The measurement is what turns "should be okay" into something you can actually show a client — which is exactly why it's worth doing.
What this means for your site
You can't easily run this whole checklist yourself, but you can run the single most important item — the raw-HTML test from the last post. If your content survives that, you've cleared the highest hurdle. The rest — landmarks, hit areas, focus, semantics — is the sort of thing I check and fix as a matter of course.
The deliverable isn't a badge or a buzzword. It's a pass/fail you can trust, on a standard Google actually published, for a site that AI assistants, search engines and every one of your visitors can genuinely use.
If you'd like your own site run through this checklist, I offer it as a fixed-price AI & agent readiness audit — or see how I build.