The three eras (Raymond’s diagram)
“LLM turns a user query into fan-out queries which gets sent to search engines.”In the “Past” world you optimized for the human’s exact phrasing and fought for position #1. In the “Now” and “Next” worlds an LLM sits in the middle. It reformulates the human’s intent into many queries, reads a handful of pages, and cites a few. Ranking #1 for the original phrasing is neither necessary nor sufficient, because what you actually want is to be citable across the whole fan-out. You don’t have to guess at that fan-out. You can watch the machine’s actual queries. Synscribe ships tools for exactly this: a Chrome extension that reveals ChatGPT’s web-search queries, and Birdseye (a macOS app) that shows Claude Code’s queries. It’s the same move you’d make as a human: step into the shoes of a serious customer researching the space, and ask how they would find your product.
What machines actually cite
The GEO content strategy comes straight from citation-type data (“What ChatGPT Gobbles Up”, promptwatch.com):
This hierarchy is why our content mix leans on landing pages (see
1.6) and starts BOFU-first (see 1.3): it
maps to what the machines pull from. Real citations already work this way in the wild.
best duty drawback pulls Zollback, hr intake automation pulls Jinba, how to set reminder linkedin dm pulls Kondo, and assent vs comply pro pulls Reglyr. Each one is a specific,
high-intent query answered by a matched page.
❓ [needs Raymond: referral tracking]. PLAN 1.2 says “why we track ChatGPT/Perplexity/Claude/ Gemini referrals,” but the Workshop Notes cover seeing queries (extension/Birdseye), not the referral-tracking rationale. Confirm the tracking philosophy (likely lives in the PostHog/Onboarding SOP).
Operationalize it
- Fan-out thinking when you find keywords: 2.2.1 — Keyword exploration.
- See and measure AI-engine traffic separately: 2.6.1 — PostHog dashboard setup and 1.10 — Attribution.
- Why the agent-in-the-middle changes how we operate, too: 1.12 — Agent-operated SEO.