Cloudflare Worker skill

LLM Cloaking Worker

Show different content to ChatGPT, Claude, and Perplexity than you show to humans and Google. A user-agent gated content switch that runs at the Cloudflare edge.

Skill source on GitHubSource article: iPullRank

What it does, in one line

A Cloudflare Worker sits in front of your site, reads the User-Agent header on every request, and routes LLM crawlers down a different path than humans and search engines.

The request flow

  1. 1

    Request hits Cloudflare edge

    Browser, bot, or LLM crawler issues a GET. Cloudflare routes it to the Worker before it touches your origin.

  2. 2

    Worker reads the User-Agent

    Pattern matches against the LLM crawler block list. Real browsers, Googlebot, and Bingbot pass through untouched.

  3. 3

    One of two modes fires for LLM UAs

    protect returns 404 for protected JS bundles, leaving the bot with an empty HTML shell. translate rewrites the URL to a markdown twin and serves that instead.

  4. 4

    Humans and Google see full content

    Normal users render the full app. SEO ranking is unaffected because real Googlebot is allow-listed and gets the canonical HTML.

Two modes

protect

Returns 404 for protected JavaScript bundles when the requester is an LLM crawler. The bot gets the empty HTML container and nothing else. About 69 percent of LLM crawlers cannot execute JavaScript, so they leave with no content.

Best for: proprietary content gated behind client-rendered JS. Audits, dashboards, datasets, paid frameworks.

translate

Rewrites the request to fetch a markdown twin of the page (for example, /about.md instead of /about) and serves that to LLMs. They still get content, but only the version you wrote.

Best for: public content you want cited in AI Overviews and ChatGPT search, with a curated summary and roughly 80 percent lower token cost.

User-agent rules

Block these

LLM crawlers. Refresh monthly against the https://github.com/ai-robots-txt/ai.robots.txt list.

  • GPTBot
  • ChatGPT-User
  • OAI-SearchBot
  • ClaudeBot
  • anthropic-ai
  • Claude-Web
  • PerplexityBot
  • Perplexity-User
  • CCBot
  • Bytespider
  • Meta-ExternalAgent
  • Google-Extended
  • Applebot-Extended
  • Amazonbot
  • cohere-ai

Never block these

Real search engines. Blocking them tanks rankings. Note thatGoogle-Extended and Applebot-Extended are the AI-training agents, not the search bots, and are safe to block.

  • Googlebot
  • Googlebot-Image
  • Googlebot-News
  • Bingbot

When to use

Use it for

  • Proprietary content being harvested by LLM crawlers
  • Audits, datasets, frameworks, surveys, benchmarks
  • Pages where you want a controlled markdown summary cited
  • Sites already behind Cloudflare

Do not use it for

  • Top-of-funnel SEO content that needs AI citation
  • Sites not behind Cloudflare
  • Public marketing pages you want LLMs to amplify
  • Blocking real search engines, ever

Quick start

  1. 1. Run the llm-cloaking-worker skill in Claude Code. It generates a wrangler project.
  2. 2. Drop in the included worker-template.ts and pick a mode (protect, translate, or both).
  3. 3. Set the route in wrangler.toml to your Cloudflare-fronted domain.
  4. 4. Deploy with npx wrangler deploy.
  5. 5. Validate with the included curl harness in test/harness.mjs spoofing GPTBot vs a regular browser UA.
Open the skill on GitHub