Methodology

We measure technical readiness, not AI mentions: we run 27 checks across 6 layers, applied to what can be observed publicly. We do not run an AI agent on the site; simulating actions is a separate module, in development, with its own score. Each check gives OK (1), Partial (0.5) or Missing (0); checks that do not apply to the site type (N/A) are left out of the calculation. A layer’s score is the weighted average of its checks, and the total score is the weighted average of the layers. Registry version: 0.1.0.

Grades

A 90–100

B 75–89

C 60–74

D 40–59

F below 40

Layers and checks

Discovery

weight 20%

robots.txt: rules for AI crawlers
HTTP parity per AI user-agent (no WAF blocking)
XML sitemap includes vehicle/product pages
/llms.txt present and structured
Content visible without JavaScript
/.well-known/api-catalog (RFC 9727)
Performance: TTFB and HTML weight

Identity

weight 15%

JSON-LD AutoDealer / AutomotiveBusiness / Organization
LocalBusiness: address, geo, opening hours, phone
Brands represented (brand) + locations (department)
NAP consistency (phone) across pages and schema
sameAs links (Google Business Profile, social)
Language declared correctly (lang / hreflang)

Inventory

weight 20%

Vehicle/product pages with a unique URL
Complete Car/Vehicle/Product + Offer JSON-LD
JSON-LD price = displayed price
Public stock/product feed (XML/JSON/CSV)
Crawlable search/filter URLs

Actions / WebMCP

weight 20%

WebMCP: registered tools (declarative / imperative)
Expected tools for this site type
Tool quality: verb_object names, descriptions, parameters
Semantic lead forms (label, name, autocomplete), not CAPTCHA-only
Public MCP endpoint (marketplace / importer)

Trust

weight 10%

Write tools with confirmation (no autosubmit on POST)
No AI-directed instruction text in the page / metadata
HTTPS + redirect; cookie consent that does not block content
Offers with validity marked (priceValidUntil)

Agent UX

weight 15%

3 intents (e.g. “request a quote”), evaluated from the results of the other checks.

Goals

The same checks, grouped by practical questions. A check can appear under several goals.

1. Can an AI agent discover the dealer?
2. Does the site let AI agents in?
3. Does the agent understand who the dealer is?
4. Can the agent read the inventory?
5. Can the agent get a trustworthy price? — Coming soon: detailed final price (taxes, additional costs).
6. Can the agent take action?
7. Is the agent integration safe and well built?
8. Can the agent submit a lead with consent? — Coming soon: lead submission with consent and an audit log.
9. Can the agent use the site directly?