paulmiller.com — AI Site Grade

Analysis

Anthropic-Exclusive Crawler Access

The site maintains a reverse bot wall: anthropic-ai receives a 200 response with 376KB of full HTML content, while every other AI crawler tested — GPTBot, Google-Extended, ClaudeBot, PerplexityBot, OAI-SearchBot, ChatGPT-User, Applebot-Extended, Bytespider — gets a 403 Access Denied from AkamaiGHost. Browser user-agents also receive 403. The robots.txt and llms.txt files are inaccessible (403) to all user-agents. The sitemap at https://paulmiller.com/sitemap.xml (589KB, 1,750+ URLs) is only readable by anthropic-ai via nginx, bypassing the Akamai layer. This creates a situation where only Anthropic's Claude can index the site's content, while Google, OpenAI, Perplexity, and Apple are entirely locked out.

Cold-Knowledge Gap

LLM knowledge about Paul Miller Auto Group describes a multi-franchise New Jersey dealership group (Ford, Kia, Mazda, Nissan, VW, Porsche, Audi, BMW, Subaru, Toyota, Rolls-Royce, Bentley, Honda, Chevrolet) with a "Paul Miller Lifetime Warranty" and 12 showrooms serving Parsippany/Wayne since 1976. The Wayback Machine snapshot from November 2024 confirms this: the homepage promotes 12 brands, the "Paul Miller Difference" philosophy, Fastlane online buying, and a Certified Pre-Owned program. However, the live site's homepage HTML (fetched as anthropic-ai) contains CSS classes referencing all those brands plus Rolls-Royce and Bentley — but the actual visible text content is entirely JS-rendered and not extractable from the raw HTML. The site is a single-page application shell (dealer.com DDC platform) where inventory and text load dynamically. The cold knowledge is richer than what the raw HTML delivers to crawlers.

Schema and Structured Data Posture

The homepage HTML contains zero JSON-LD schema of any type. No AutoDealer, Organization, LocalBusiness, Product, or Vehicle schema is present in the source. The meta tags include robots: index, follow, og:title, og:type, og:url, and locale: en_US, but no structured data that would help AI models understand the dealership's inventory, location, brands, or services. The ddc-site platform typically injects schema via JavaScript, meaning only JS-capable crawlers (none of which pass the Akamai wall except anthropic-ai) would ever see it.

External Signal Fragmentation

DuckDuckGo returns zero search results for "paulmiller.com", "Paul Miller Auto Group", or any combination of the dealership name with New Jersey — a complete absence from the search index. The domain has 1,750 Wayback Machine captures since 1996, indicating long operational history, but the current Akamai configuration blocks all search engine crawlers. The DNS points to dealer.com nameservers and Akamai IPs, with Microsoft 365 for email. The site's external reputation (reviews, press, Reddit) is entirely invisible to search engines because no crawler can access the content to index it.

Findings

1. Akamai firewall blocks all AI crawlers except Anthropic's Claude High

   The site returns 403 Access Denied to GPTBot, Google-Extended, ClaudeBot, PerplexityBot, OAI-SearchBot, ChatGPT-User, Applebot-Extended, and Bytespider. Only anthropic-ai receives a 200 response with full HTML content. This creates an exclusive indexing privilege for Anthropic while locking out all other AI platforms.

   What to change: Remove the Akamai bot-blocking rules that deny non-Anthropic AI crawlers, or replace with a permissive robots.txt and allow all legitimate crawlers.

2. robots.txt and llms.txt return 403 to all user-agents High

   Both robots.txt and llms.txt are inaccessible (403) to all user-agents, including anthropic-ai. This prevents crawlers from discovering allowed paths and signals poor crawl governance.

   What to change: Serve a publicly accessible robots.txt that allows all well-behaved crawlers and a llms.txt that lists key pages.

3. Sitemap is only readable by Anthropic's Claude via nginx bypass High

   The sitemap at https://paulmiller.com/sitemap.xml (589KB, 1,750+ URLs) returns 200 only to anthropic-ai, which accesses it via nginx bypassing the Akamai layer. All other crawlers receive 403.

   What to change: Make the sitemap publicly accessible to all crawlers by removing the Akamai restriction.

4. Homepage is a JS-rendered SPA with no extractable text content High

   The homepage HTML (fetched as anthropic-ai) contains CSS classes referencing 12+ car brands but the visible text is entirely JavaScript-rendered and not present in the raw HTML. Crawlers that cannot execute JavaScript see an empty shell.

   What to change: Implement server-side rendering or pre-rendering for key pages so that static HTML includes meaningful text content.

5. Homepage contains no JSON-LD structured data High

   The homepage HTML has zero JSON-LD schema of any type (AutoDealer, Organization, LocalBusiness, Product, Vehicle). No structured data helps AI models understand the dealership's inventory, location, brands, or services.

   What to change: Add JSON-LD schema for AutoDealer, Organization, LocalBusiness, and Vehicle on all relevant pages.

6. Domain has zero search engine index presence High

   DuckDuckGo returns zero results for 'paulmiller.com', 'Paul Miller Auto Group', or any combination with New Jersey. The site is completely absent from search indexes due to the Akamai block.

   What to change: Remove the Akamai block on search engine crawlers and submit the sitemap to Google Search Console and Bing Webmaster Tools.

7. External reputation signals are invisible to search engines Medium

   Reviews, press mentions, and Reddit discussions about Paul Miller Auto Group are not indexed because no crawler can access the site content to associate with the domain.

   What to change: Allow search engine crawlers to access the site so that external signals can be associated with the domain.

8. Structured data is injected via JavaScript and invisible to non-JS crawlers Medium

   The DDC platform typically injects schema via JavaScript, meaning only JS-capable crawlers (none except anthropic-ai pass the Akamai wall) would ever see it. Non-JS crawlers see zero schema.

   What to change: Include JSON-LD schema in the server-rendered HTML rather than injecting via JavaScript.

9. LLM cold knowledge is richer than live site content Medium

   LLM knowledge describes a multi-franchise dealership with 12 brands and a lifetime warranty, but the live site's raw HTML delivers no extractable text to confirm this. The cold knowledge is more detailed than what crawlers can actually retrieve.

   What to change: Ensure that key content (brands, warranty, locations) is present in server-rendered HTML so crawlers can extract it.

10. No llms.txt file available for AI crawlers Low

    The llms.txt file returns 403, preventing AI crawlers from discovering a curated list of important pages.

    What to change: Create and serve a publicly accessible llms.txt file listing key pages like inventory, about, and contact.

What's working

• Anthropic's Claude can access full HTML content — The site allows anthropic-ai to fetch the homepage (376KB) and sitemap (589KB) with full HTML, enabling Claude to index the site's content.
• Sitemap contains 1,750+ URLs for comprehensive indexing — The sitemap at /sitemap.xml lists over 1,750 URLs, providing a complete inventory of pages for crawlers that can access it.
• Domain has 1,750+ Wayback Machine captures since 1996 — The domain has a long operational history with extensive archival records, indicating established online presence.
• Basic meta tags (og, robots) are present in HTML — The homepage includes og:title, og:type, og:url, locale, and robots meta tags, providing basic metadata for social sharing and crawl directives.