Skip to content

AI-Compatible v3

Repo: savvydealer-adam/ai-compatible-v3 · Path: C:/Users/adam/ai-compatible-v3 · Owner: Adam Status: Live · % Done: 90 · Last commit: 2026-04-08 Deployed: Cloud Run in GCP project aicompatibility (NOT savvydealer-website)

What it is

AI-compatibility audit for dealer websites. Answers two strictly separated questions for any dealer URL: AI Access (can AI tools reach the site?) and JSON-LD Structured Data (does the HTML literally contain <script type="application/ld+json"> with the expected schema.org types?).

Why it exists

AI search (ChatGPT, Perplexity, Gemini) is eating traditional dealer SEO. Dealers have no visibility into whether AI assistants can actually see their inventory, pricing, and hours. v3 answers that with the two-dimension split that earlier versions kept conflating — never let "AI says it sees the inventory" be mistaken for "the HTML has JSON-LD."

How it works

Stack: Python + FastAPI, React + Vite client, Cloud Run in the aicompatibility GCP project. - server/ — FastAPI backend; scoring logic split into AI Access vs JSON-LD modules - client/ — React + Vite frontend - AI Access path — probes robots.txt, llms.txt, crawler access, and live-tests with 4 AI providers (blocked-fallback strategy if one provider is restricted) - JSON-LD path — fetches raw HTML, parses <script type="application/ld+json"> tags, inspects @type (AutoDealer/LocalBusiness for home, ItemList/OfferCatalog for SRP, Car/Vehicle/Product for VDP). The only way to detect JSON-LD. Never inferred from AI behavior, sitemaps, or robots.txt. - PLAN-ai-live-testing.md documents the AI live-testing strategy - Deploy via cloudbuild.yaml

What's done

  • Strict two-dimension separation (AI Access vs JSON-LD) — codified in CLAUDE.md
  • JSON-LD parser validates @type, required fields (name, address, telephone, openingHoursSpecification, offers.price, VIN, etc.)
  • 4-provider AI live testing with blocked-fallback
  • Frontend dashboard showing both scores separately, never collapsed
  • Deployed to Cloud Run

What's next

  • Extend VDP JSON-LD validation (more edge cases — UsedCar, MotorVehicleSpec, etc.)
  • Self-audit SavvyDealer-hosted dealer sites on a cron and publish a scoreboard
  • Tighter wiring to cloud-savvydealer-website so prospects can run an audit inline on the marketing site

Where the code lives

  • FastAPI entry: server/ main module
  • Client entry: client/src/App.tsx
  • Scoring docs: CLAUDE.md (read before touching any scoring code), PLAN-ai-live-testing.md
  • Tests: tests/, test_ai_verify.py
  • Deploy: cloudbuild.yaml, GCP project aicompatibility

Integrations

  • 4 AI providers — used to live-test whether AI can reach the site
  • dealer websites / WP Migration Factory — target audit surface
  • Successor to legacy ai-visibility (superseded)

Don't rebuild this — extend it

This is THE authoritative scorer for AI-readiness. Do not reinvent JSON-LD detection elsewhere — call this service or extend server/ scoring modules. Never collapse AI Access and JSON-LD into a single number.