An agent-ready website is one an AI agent can parse, trust, and eventually transact with, and the version of that standard that matches the evidence comes in phases, not a single bar. Google's own guidance now describes agents browsing sites through the DOM and accessibility tree and points to emerging commerce protocols; the preparation is real work. But the most common failure we see isn't being unprepared. It's buying phase-three deliverables for a site that still fails phase one.
So here's the full checklist, phased, with the placement test built in. Run it top to bottom; where you stop passing is the phase you're in, and the only phase worth spending on.
Phase 1: fundamentals (everyone, now)
This phase serves three audiences at once (humans, AI answers grounding on your pages, and browser agents parsing them), which is why it's never wasted regardless of how the protocol layer shakes out.
Indexable and crawlable, verified rather than assumed: no leftover noindex, no blocking robots.txt, a submitted sitemap. Non-commodity content on the pages that matter, because Google's guidance is explicit that unique, expert content is what its AI features ground on. Semantic HTML: real headings in a sensible hierarchy, labeled form fields, actual buttons, lists and tables marked up as what they are. Speed measured, not felt: first byte under half a second is the working bar. And structured data where it demonstrably serves a surface (business info, products, articles), kept truthful.
Phase 2: agent readiness (data-rich businesses, this year)
This phase makes your business facts machine-consumable and your reliability visible.
Your first-party facts (services, prices, hours, availability, service area) live on real pages as real text, consistent with your Business Profile, per the source-of-truth inventory. The accessibility tree of your key pages is legible on its own: open DevTools, read the tree, and ask whether a machine could tell what you offer and how to act. Uptime is monitored on the paths that matter (home, contact, booking), because an agent mid-handoff treats downtime as your failure. Agent traffic is logged: grep for Google-Agent, baseline the count, revisit monthly. And anything endpoint-shaped you already expose uses modern authentication.
Phase 3: agentic presence (transactable businesses, when the ecosystem justifies it)
The forward phase, and the one to enter deliberately: an Agent Card drafted at /.well-known/agent.json, capability design (deciding what your brand's agent should answer and do), commerce-protocol readiness (UCP and WebMCP watching, clean feeds, checkout API posture), and the new KPI layer instrumented as the events start existing: task handoffs, invocations, agent-referred conversions.
The checklist, in full
Yours to keep, whether you ever hire us or not:
PHASE 1: FUNDAMENTALS
[ ] No stray noindex; robots.txt clean; sitemap submitted
[ ] Money pages: non-commodity content, answer-first structure
[ ] Semantic HTML: real heading hierarchy, labeled forms, true buttons
[ ] First byte < ~0.5s on key pages (measure with curl, not vibes)
[ ] Structured data where it serves a surface, and truthful
PHASE 2: AGENT READINESS
[ ] Services, prices, hours, availability: on-page, text, consistent
with your Business Profile
[ ] Accessibility tree legible on key pages (DevTools check)
[ ] Uptime monitor on home + contact/booking paths
[ ] Server logs checked for "Google-Agent"; baseline recorded
[ ] Modern auth (OAuth 2.0-family) on any exposed endpoints
PHASE 3: AGENTIC PRESENCE
[ ] Agent Card drafted at /.well-known/agent.json
[ ] Capability design: what your agent answers, what it can do
[ ] Commerce readiness: product feeds clean, UCP/WebMCP watched
[ ] Agentic KPIs instrumented as events exist (handoffs, invocations)
The failure we see most, per phase
Having run this against real sites, each phase has a signature miss worth naming, because they're predictable enough to check first.
Phase 1's is the page-builder tax: sites that look immaculate and parse as div soup, because the theme renders headings as styled paragraphs and buttons as clickable divs. The human eye can't see it, which is exactly the problem; the accessibility-tree check exposes it in one look.
Phase 2's is fact drift: the machine-readable surfaces exist but disagree. Hours differ between the site and the Business Profile, a service is listed in one place and priced in another, and an agent cross-checking sources finds a business that contradicts itself. Consistency is the phase-2 skill more than coverage.
Phase 3's is buying it early. Not a technical failure at all: a budget one. An Agent Card wired to stale data is worse than no card, because it converts your data problems into confidently wrong machine-readable answers, delivered at scale, with your name on them.
The placement test, and the candid part
Score each phase separately. Passing means every box, because agents don't average: one unparseable booking form undoes five fast pages.
The candid part is where most readers land. Phase 3 is premature for almost everyone reading this; the ecosystem is still enterprise pilots, and a phase-3 invoice sold to a phase-1 site is the "AI SEO" industry's favorite margin. The checklist exists so you can name your phase before someone else names it for you. Phase 1 gaps: fix them this month, they pay immediately. Phase 2 gaps with a data-rich business: this year's project. Phase 3: watch, prepare cheaply, enter when the task-handoff metrics have events to count.
We run these same phases as the Reforge Method's spine, and for enterprise teams that need the full agent-readiness layer designed and built, that's what Frontier Advisory is for. For everyone else: the checklist is the deliverable, and it's already yours.