Cloudflare Worker

Fresh 2026-06-03

File: cloudflare-worker.jsTarget route: andersonhemmat.com/practice-areas/car-accident-lawyers*Status: Code complete — awaiting Cloudflare DNS setup

---

What It Does

The Worker intercepts incoming requests to the car accident page and:

1. Classifies the requester as bot or human using multi-signal detection
2. Humans always receive the normal origin page — zero impact on human visitors
3. Identified AI bots receive an AI-optimized content variant from KV storage
4. All bot visits are logged to Cloudflare KV for monitoring

---

Detection Architecture

flowchart TD
    A[Incoming Request] --> B{Is GET to target path?}
    B -->|No| Z[Pass to origin unchanged]
    B -->|Yes| C[Run classifyRequest]
    C --> D{Signal 1: UA match?}
    D -->|Yes| E[High confidence bot]
    D -->|No| F{Signal 2: Bot score < 30?}
    F -->|Yes| G{Signal 3: ASN in AI list?}
    G -->|Yes| H[Medium confidence bot]
    G -->|No| I[Low confidence bot]
    F -->|No| J{Signal 4: DeepSeek spoof?}
    J -->|Yes| E
    J -->|No| K[Human — pass to origin]
    E --> L[selectVariant - high/med]
    H --> L
    I --> M[selectVariant - low, default md]
    L --> N[Fetch from KV]
    M --> N
    N --> O{Variant in KV?}
    O -->|Yes| P[Serve variant with headers]
    O -->|No| Z
    P --> Q[Log to AI_VISITS_LOG async]

---

Multi-Signal Classification

The Worker uses 4 independent signals, combining with AND/OR logic:

Signal 1: User-Agent Pattern Matching (Highest Specificity)

27+ patterns covering:

• OpenAI: GPTBot, ChatGPT-User, OAI-SearchBot
• Anthropic: anthropic-ai, ClaudeBot, Claude-Web
• Google: Google-Extended, Gemini, Googlebot-AI
• Perplexity: PerplexityBot
• Meta: FacebookBot, Meta-ExternalAgent
• Microsoft: bingbot, BingPreview
• Apple: Applebot-Extended
• Cohere, Amazon, ByteDance (Bytespider), and 10+ others

Signal 2: Cloudflare Bot Score

cf.botManagement.score < 30 = bot-likely. Requires Cloudflare Bot Management plan.

Signal 3: ASN (Autonomous System Number)

Known AI lab datacenter IP ranges:

• AS8075 (Microsoft/OpenAI), AS16509/AS14618 (AWS/Anthropic), AS15169/AS396982 (Google), AS32934 (Meta), AS138699/AS62567 (ByteDance/DeepSeek), AS14061 (DigitalOcean), AS24940 (Hetzner)

Signal 4: DeepSeek Spoof Detection

DeepSeek spoofs Chrome 58 User-Agent but uses ByteDance/Chinese ASNs. Detection logic:

function isDeepSeekSpoof(request, asn) {
  const ua = request.headers.get("User-Agent") || "";
  const isFakeChrome58 = /Chrome\/58/.test(ua) && /Mozilla\/5\.0/.test(ua);
  const deepseekASNs = new Set(["AS138699", "AS62567", "AS63086", "AS45090", "AS4134", "AS4837", "AS9808"]);
  return isFakeChrome58 && deepseekASNs.has(asn);
}

---

Confidence Levels

Confidence	Trigger	Action
High	UA match OR DeepSeek spoof	Serve variant directly
Medium	Bot score low AND ASN match	Serve variant directly
Low	Bot score low only	Serve variant (default: md)
None	No signals	Pass to origin unchanged

---

Variant Selection

Bot Type	Variant	Reason
openai, anthropic, google, perplexity, microsoft, cohere, amazon, apple	md	LLM assistants parse Markdown best
bytedance, deepseek	schema	Entity authority focus
ai_assistant_unidentified, ai_crawler_other	html	Universal fallback
low confidence	md	Safest option

---

Response Headers

All variant responses include these headers:

X-Served-By: anderson-hemmat-ai-worker
X-Variant: md|html|schema
X-Bot-Type: openai|anthropic|...
X-Bot-Confidence: high|medium|low
X-Bot-Signals: ua:GPTBot|bot_score:12|...
Cache-Control: no-store
X-Robots-Tag: noindex

noindex Header

The X-Robots-Tag: noindex header prevents AI variants from being indexed by Google as separate pages. The variants are for AI assistant consumption only, not for search indexing.

---

Logging

All bot visits are logged to KV namespace AI_VISITS_LOG:

{
  "timestamp": "2026-06-03T12:00:00.000Z",
  "url": "https://andersonhemmat.com/practice-areas/car-accident-lawyers",
  "botType": "openai",
  "confidence": "high",
  "signals": "ua:GPTBot",
  "botScore": 2,
  "asn": "AS8075",
  "variantServed": "md",
  "userAgent": "GPTBot/1.0",
  "country": "US",
  "ip": "192.0.2.1"
}

Log entries have 90-day TTL. Query examples in the Measurement Framework.

---

Worker Bindings Required

name = "anderson-hemmat-ai-variants"
main = "cloudflare-worker.js"
compatibility_date = "2026-01-01"

[[routes]]
pattern = "andersonhemmat.com/practice-areas/car-accident-lawyers*"
zone_name = "andersonhemmat.com"

[[kv_namespaces]]
binding = "AI_VARIANTS"
id = "<KV_NAMESPACE_ID>"

[[kv_namespaces]]
binding = "AI_VISITS_LOG"
id = "<KV_LOG_NAMESPACE_ID>"

---

Human Traffic: Zero Impact

The Worker passes human traffic through to origin unchanged on two conditions:

1. classification.isBot === false — no bot signals detected
2. request.method !== 'GET' — non-GET requests (forms, AJAX, etc.) always pass through

Human visitors never see the AI variants. Google's main crawler (Googlebot) also passes through — only Google-Extended (the AI training crawler) receives the variant.