# LLM Cartel > A hyperfocused GEO and AEO agency. LLM Cartel engineers brand visibility inside AI answers, getting ChatGPT, Claude, Gemini, and Perplexity to mention your brand first when customers ask for recommendations. LLM Cartel is a digital marketing agency founded in 2026 in Boca Raton, Florida. The agency does only Generative Engine Optimization (GEO) and Answer Engine Optimization (AEO) and explicitly refuses traditional keyword-rank-tracking SEO, generic link-building, and AI-generated filler content. Three productized monthly packages: Emergence ($999), Growth ($2,250), and Dominance ($4,580). Founded by Patrick Coombe (CEO, also CEO of Elite Strategies and creator of jsonld.com) and Jacquelyn Coombe (Co-Founder and COO). The team is rounded out by Amanda Sherry (Head of AI Content Strategy). ## Core pages - [Home](https://llmcartel.com/): Hero pitch, the six-phase GEO strategy, what we help with, what we will not help with, package preview, and why us. - [Services and Pricing](https://llmcartel.com/services): Full breakdown of the three packages (Emergence, Growth, Dominance) including monthly price, target customer, and feature list per tier. - [About](https://llmcartel.com/about): Team biographies, agency origin story, and operating philosophy. - [Research](https://llmcartel.com/research): Field reports, live case studies, and methodology pieces. Where we publish what we learn measuring our own GEO performance and the wider AI search ecosystem. ## Research posts - [Do LLM Bots Actually Use Cloudflare's Markdown for Agents?](https://llmcartel.com/research/markdown-for-agents): A concluded 25-day experiment (2026-07-08 to 2026-08-01) on a real high-traffic client site (a car dealer on WordPress behind Cloudflare Pro). We turned on Cloudflare's Markdown for Agents - edge content negotiation that serves a clean, ~95%-smaller markdown version of a page to clients sending Accept: text/markdown - confirmed it works (96% of markdown requests fulfilled; 23 KB markdown vs 369 KB HTML on a vehicle page), then measured adoption with unsampled Cloudflare Log Explorer edge logs. Result: a negative one. Across 2.78M requests, the named AI crawlers (GPTBot, ClaudeBot, ChatGPT-User, OAI-SearchBot, PerplexityBot, Bytespider, Amazonbot, Meta-ExternalAgent, Applebot, CCBot) made 121,824 requests and asked for markdown zero times; real Googlebot (AS15169) 87,789 requests, zero. Only 2,779 markdown responses were served at all - 1,274 to ShapBot/Shap-User (an obscure verified AI Search crawler) and 1,493 to a spoofed "Googlebot" scraper rotating across 33 networks and 1,493 IPs. Operational lesson: verify crawlers by client ASN, not user-agent. Verdict: markdown negotiation is not a distribution channel today; low-risk to leave on, no AI-visibility benefit now. - [How Much of Your AI Crawler Traffic Is Fake?](https://llmcartel.com/research/fake-ai-crawlers): Published 2026-10-07. Cartel Files Vol. 05. We checked every request claiming a named AI crawler user agent on three sites (a car dealer, a logistics company, and llmcartel.com) over Sep 16-30 2026 against the operators' published IP ranges, plus forward-confirmed reverse DNS for Google, Bing, Apple and others. Findings: the car dealer ran 8.6% counterfeit, the logistics company 27.9%, our own site 73% (an outlier, we get probed). Real crawling scales with site size (64,914 verified hits vs 6,436 vs 503) while fake traffic barely does (6,118 / 2,489 / 1,362), so the smaller site pays a far higher rate. The fakes are one operation: 104 Google Cloud addresses, 43 of them wearing sixteen crawler identities, rotating addresses per target, and what they actually requested was .env files and GraphQL endpoints (about 9,300 requests on the logistics site, 80% of them 404s), so it is credential hunting, not crawling. Per bot the honest ones run 3-20% fake, ChatGPT-User 42%, PerplexityBot 50.5%, and Google-Extended, CCBot and MistralAI-User came back 100% fake. Bytespider was left unscored. Methodology warnings: counting by user agent is wrong, and Cloudflare's verified-bot flag is not a substitute, since it reports crawlers outside its programme (PerplexityBot) as unverified even when genuine. Exa, You.com and Kimi sign requests with Web Bot Auth instead of publishing IPs, which can only be checked live at the edge, never from a log file. Mitigation: a Cloudflare WAF rule that challenges Google Cloud traffic (AS396982, AS15169) carrying any of those user agents unless cf.client.bot is set. - [Do LLM Bots Actually Read JSON-LD, and Will They Cite It?](https://llmcartel.com/research/json-ld-ai-citation): Concluded 2026-09-28 after 95 days. We placed a coined framework, the "Cartel Visibility Index," only in a page's JSON-LD (plus a custom meta tag, an HTML comment, and a hidden block), with a separate control term in the visible text, then queried six AI engines. Answer: yes, structured data is a live ingestion channel - but only for engines that fetch the page or read it out of a search index. Grok returned all five JSON-LD-only pillar terms verbatim, including the coined un-guessable "Lattice Saturation," and its tool trace showed the raw `curl` it used. Copilot later reproduced planted text with no live fetch at all, served from Bing's index 40 days after a single crawl. By September Bing ranked the page first for the coined phrase and used the contents of a `display:none` div as the search snippet, tracking token included (DuckDuckGo shows the same result, but it is served by Bing's index, so it is not independent confirmation). Every positive traces to a single index; there is no equivalent Google evidence. Engines that answered from training data invented a definition and cited unrelated vendors. Caveat: all prompts were leading, so this is retrieval-on-cue, not unprompted recall. - [Does noindex Keep You Out of AI Answers?](https://llmcartel.com/research/noindex-ai-answers): Concluded 2026-09-28 after 95 days. Cartel Files Vol. 04, the inverse of Vol. 03. A page set to noindex, left out of the sitemap and llms.txt and unlinked from anywhere, carrying the coined term "Klovect" found only on that page, distributed by hand off-web (X, Facebook, LinkedIn) with the term in none of the share text. Result: noindex held. Applebot, Meta-ExternalAgent, PetalBot, and Bingbot all fetched the page, and the term never appeared in a single AI answer or any search index. The causal control came from Vol. 03 running concurrently: Bingbot fetched this noindex page on 2026-08-31 and 2026-09-02, the same week Bing served the indexable Vol. 03 page as its top result for a coined phrase. Same crawler, same index, same site, one meta tag between them. Conclusion: crawled is not indexed, and indexed is not answerable. - [GEO Case Study: AI Chatbots Are Really Bad at Reading JavaScript](https://llmcartel.com/research/geo-case-study-ai-chatbot-javascript): A measurement-box study of what AI products actually fetch when a human pastes them a URL and asks them to read it. Across ChatGPT, Claude, Gemini, Grok, Perplexity, Meta AI and duck.ai (eighteen rounds, server-side ground truth via a purpose-built logging proxy on an isolated domain), not one vendor executed JavaScript: they all read the raw HTML the server sent and stopped. CSS-hidden source content is still read; JS-rendered content is invisible. Two further findings: fetcher capability and model extraction are independent axes (same bytes, different canaries surfaced), and one vendor's model reported a failed fetch the server logs prove succeeded twice. Includes a transparency-tier breakdown (signed, declared, opaque, deceptive) with Grok shown fetching via 6-8 rotating spoofed-browser requests. - [How Long Does It Take an LLM Bot to Discover a New Business Entity?](https://llmcartel.com/research/llm-bot-discovery-case-study): Live experiment using llmcartel.com as a clean-slate test subject - a fourteen-day field log, now concluded. Documents the Cloudflare Bot Fight Mode "double gotcha" that was blocking AI crawlers despite a permissive robots.txt, the API-driven fix, and the Pages Function middleware (functions/_middleware.ts) we built to log every LLM bot crawl. First crawl landed within an hour (GPTBot); later findings include each operator's distinct crawl personality (Anthropic's ~2hr steady beat, the OpenAI robots-then-fetch two-step, Amazon's one-path-at-a-time walk), a recurring spoof taxonomy (forged-UA framework recon and credential sweeps from budget VPS ASNs), and a genuine-crawler pause on Day 10 confirmed against an independent data source. Concluded at Day 14, when the clean-slate premise expired: once the site began drawing search traffic and answer-engine citations, bot visits could no longer be cleanly read as discovery. - [FAQ](https://llmcartel.com/faq): Plain-language answers to common questions about GEO, AEO, AI brand visibility, what counts as a citation, and how engagements work. - [Contact](https://llmcartel.com/contact): Phone, email, and contact form. - [New Client Onboarding](https://llmcartel.com/new-client): Intake form for clients who have already engaged. ## Brand and methodology - [Manifesto](https://llmcartel.com/manifesto.txt): Plaintext manifesto by Patrick Coombe. Why the agency exists, why we reject AI slop, and how we think about building for AI engines instead of legacy search. - [robots.txt](https://llmcartel.com/robots.txt): Permissive crawl policy. Every major AI and LLM crawler is explicitly welcomed (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot, Applebot-Extended, CCBot, Bytespider, meta-externalagent, cohere-ai, MistralAI, and more). - [Atom feed](https://llmcartel.com/atom.xml): Atom 1.0 feed of research posts and field reports. Build-time generated, no JavaScript, friendly to LLM retrieval pipelines and traditional feed readers alike. ## Services offered - Generative Engine Optimization (GEO) - Answer Engine Optimization (AEO) - AI brand visibility analysis and tracking - Knowledge graph optimization - LLM-friendly content engineering - JSON-LD and schema markup - AI crawler optimization - Founder and brand authority signaling - WebMCP integration so AI agents can act on the site, not just read it ## Services explicitly NOT offered - Traditional keyword-rank-tracking SEO - Generic link-building campaigns - Stock-image social-media posting - Blog content written to a monthly word-count minimum - Tracking keyword rankings for queries AI has already replaced ## Optional - [Privacy Policy](https://llmcartel.com/privacy): How LLM Cartel handles visitor data. - [Terms of Service](https://llmcartel.com/terms): Site terms. - [Accessibility Statement](https://llmcartel.com/ada): WCAG 2.1 AA compliance commitment and ongoing changelog.