lewissilkin.com — AI Site Grade

Analysis

The critical finding is clear: lewissilkin.com is a Next.js site on Vercel that serves a 718KB JavaScript shell to all AI crawlers, returning essentially zero visible text on most pages. The homepage yields 15 words. The insights page yields 2 words. The news page yields 2 words. This is a severe AI-visibility problem.

Let me write the audit.

Lewis Silkin — AI-Visibility Audit

The entire lewissilkin.com domain is functionally invisible to AI crawlers despite returning HTTP 200 to every bot tested. The site is a Next.js application hosted on Vercel that serves a ~718KB JavaScript shell with virtually zero plain-text content — the homepage yields 15 visible words, the insights page yields 2 words, and the news page yields 2 words from a plain GET. Every AI crawler (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot, Applebot-Extended, Bytespider, anthropic-ai, ChatGPT-User, Perplexity-User) receives the same empty shell with identical byte size, meaning none are blocked — but none can extract substantive content either.

Crawler Access

The robots.txt is a bare-bones catch-all: User-agent: * Allow: / with no AI-bot-specific rules whatsoever. No GPTBot, no ClaudeBot, no Google-Extended — nothing. The llms.txt returns a 404 (redirected to a Next.js 404 page). All 11 bot UAs tested via compare_bot_access returned HTTP 200 with identical 718,247-byte payloads from Vercel servers, confirming the site is not UA-blocking but is universally JS-dependent. The DNS TXT records include an anthropic-domain-verification token, indicating the firm has engaged with Anthropic for some purpose, yet ClaudeBot receives the same empty shell as every other crawler.

Content & Schema Posture

The homepage schema is limited to Corporation and WebSite types — no LegalService, Organization, ProfessionalService, or FAQPage schema anywhere on the site. The contact page is the only page with a LegalService schema, but it references only the Oxford office with minimal address data. Key pages like the services directory, experts directory, and insights listing are all JS-rendered shells with 2–8 words of extractable text. The few pages that do render some content server-side (e.g., the "Protect your business and brand" page at ~360 words, the "Future of Work Hub" at ~229 words) contain substantive copy but lack any structured data beyond the bare WebSite/Corporation types. No Article, NewsArticle, FAQPage, BreadcrumbList, or Person schema was detected anywhere.

Cold-Knowledge Gap

The LLM prior on Lewis Silkin is surprisingly rich and accurate: it knows the firm as a UK-based employment and IP law specialist, names its HR Lunchtime Briefings, gig-economy cases, and the founder's role in the 1947 Town and Country Planning Act. This knowledge is far more detailed than anything the site itself communicates to crawlers. The site's tagline ("Protecting and enhancing what really matters to you — your ideas and people") and its "Ideas + People" H1 are generic and do not convey the firm's specific strengths in employment law, creative industries, or its Ius Laboris membership. The LLM knows about the firm's progressive reputation and top-tier rankings — none of which are evidenced in the site's schema or visible text.

External Signals

The sitemap contains 8,461 URLs, suggesting a large content library, but the vast majority of these pages are JS-rendered and invisible to crawlers. The site has locale variants for Hong Kong (en-HK, zh-HK), Ireland (en-IE), and China (zh-CN), each duplicating the same empty-shell problem. A test page (/testing-faw-2026) appears in the sitemap — a staging or draft page exposed to production. The DNS shows Mimecast for email, SuccessFactors for HR, and Mandrill for transactional email — a sophisticated enterprise stack backing a site that AI crawlers cannot read.

Findings

1. All pages render as empty JavaScript shells to AI crawlers High

   The site is a Next.js application on Vercel that serves a ~718KB JavaScript bundle with virtually no plain text. The homepage yields 15 words, the insights page 2 words, and the news page 2 words from a plain GET. Every AI crawler receives the same empty shell.

   What to change: Implement server-side rendering (SSR) or static generation for key pages so that AI crawlers receive meaningful HTML content. Alternatively, use prerendering services that serve static snapshots to bots.

2. llms.txt file returns 404 Medium

   The standard llms.txt file, which helps AI crawlers discover content, is missing and returns a 404 (redirected to a Next.js 404 page).

   What to change: Create an llms.txt file that lists key content pages and summaries to guide AI crawlers.

3. Robots.txt lacks AI-bot-specific rules Low

   The robots.txt is a bare-bones catch-all with no directives for GPTBot, ClaudeBot, Google-Extended, or other AI crawlers. While not blocking, it misses the opportunity to guide crawlers to important content.

   What to change: Add specific rules for AI crawlers, e.g., allowing access to key content sections and disallowing low-value pages.

4. No LegalService, Article, or FAQPage schema on key pages High

   The homepage only has Corporation and WebSite schema. The contact page has a LegalService schema but only for the Oxford office. No Article, NewsArticle, FAQPage, BreadcrumbList, or Person schema was detected anywhere, despite the site having services, experts, and insights pages.

   What to change: Add appropriate structured data (LegalService, Article, FAQPage, BreadcrumbList, Person) to relevant pages to help AI crawlers understand and cite the content.

5. LLM prior knowledge far exceeds site content Medium

   The LLM knows Lewis Silkin as a UK employment and IP law specialist with specific offerings like HR Lunchtime Briefings and Ius Laboris membership, but the site's tagline and H1 are generic. The site does not communicate its strengths in employment law or creative industries to crawlers.

   What to change: Ensure key differentiators (employment law focus, Ius Laboris membership, creative industries expertise) are prominently featured in visible text and structured data.

6. Staging or draft page exposed in sitemap Low

   The sitemap includes a URL '/testing-faw-2026' which appears to be a staging or draft page, indicating that test content is publicly accessible.

   What to change: Remove test or draft pages from the production sitemap and ensure they are not indexable.

7. Sitemap contains 8,461 URLs but most pages are invisible to crawlers High

   The sitemap lists thousands of URLs, suggesting a large content library, but the vast majority are JS-rendered and provide no extractable text to AI crawlers.

   What to change: Implement SSR or static generation for all pages listed in the sitemap to ensure crawlers can read the content.

8. Services and experts pages render as empty shells High

   The services page yields 8 words and the experts page yields 118 words, but both are primarily JS-rendered and lack substantive server-side content.

   What to change: Ensure service descriptions and expert profiles are rendered server-side with full text and structured data.

What's working

• No AI crawlers are blocked by robots.txt or server — All 11 tested AI crawlers receive HTTP 200 responses, and robots.txt has no disallow rules for any bot. The site is open to all crawlers.
• Anthropic domain verification token present — DNS TXT records include an anthropic-domain-verification token, indicating engagement with Anthropic for potential AI integration.
• A few pages contain substantive server-side text — Pages like 'Protect your business and brand' (~360 words) and 'Future of Work Hub' (~229 words) render meaningful content server-side, providing some value to crawlers.
• LegalService schema present on contact page — The contact page includes a LegalService schema with address data for the Oxford office, providing some structured data for local search.
• LLM has detailed prior knowledge of the firm — The LLM knows Lewis Silkin's specialization in employment and IP law, its HR Lunchtime Briefings, gig-economy cases, and Ius Laboris membership, which can aid in AI-generated citations even without site content.
• Sitemap covers a large number of URLs — The sitemap lists 8,461 URLs, indicating a broad content library that, once rendered server-side, could provide extensive material for AI crawlers.