Going global · Grow

Six schools of SEO and GEO

Each school lists its people, its core claim, what you can borrow, and what to check against official docs and data first. Names link to their own articles or channels.

The 6 SEO & GEO Practitioner Schools: Map & Fact-Check Guide

6 schools · 39 hands-on practitioners

SEO and GEO practitioners often sound like they contradict one another because each works on a different layer of search and AI retrieval. Mapping them across two axes — On-Site Product & Tech vs. Off-Site Consensus & Links (horizontal) and Long-Term Compounding Moat vs. Algorithm Arbitrage & Red-Team Testing (vertical) — shows what each school teaches, what to borrow, and where to cross-check against Google's official policies:

①Product-Led, Interactive Tools & Proprietary Assets

Layer 1: What to build (Capture vs Create Demand & 4–7 asset types)

Core thesis: Rejects cranking out filler articles from keyword spreadsheets ('Red Ocean SEO'). Strategically, Eli Schwartz and Tim Soulo (DISCOVERABLE) draw a hard line: SEO captures existing demand (Get Found), whereas unknown categories require social interruption (Get Noticed) and word-of-mouth assets (Get Talked About); Joker (2026-08-27) adds that while 'Keyword + Page + Content + Backlinks' still works in emerging low-competition niches, mature niches and AI Search require 4 higher layers (Entity + Topic + Retrieval + Network). Cyrus Shepard proved 92.9% of Core Update winners carry proprietary assets AI cannot fabricate and AIO drives 21%–23% of clicks; Georg Richard Aare's 410,572-claim study showed opening with first-hand experience lifts Top-3 probability by 5.5x.

What to borrow: 1. Put an interactive tool or structured table above the fold and open with first-hand sample size and findings in the first 50 words (Aare's 5.5x rule); 2. Combine at least 4 proprietary asset types per page (original stats, CERN Zenodo DOI dataset citation, comparison table, screenshots); 3. Build Eli's 'Southwest Airlines' mesh internal link graph.

Fact-check: Never scale empty programmatic shells or chase off-topic traffic. Joker warns that publishing 3,000 generic articles unrelated to your Entity Core is the fastest way to kill a domain; Eli's Panda recovery case study pruned 100+ near-duplicate auto sites down to 4 sites and reached 125% of pre-penalty traffic. Google doorway & thin content policy →

②Semantic Ontology, Entity Home Disambiguation & Technical SEO

Layer 2: How machines parse you (Connected @id KG +30% & RLM-on-KG)

Core thesis: Treats a website as a structured knowledge graph to eliminate machine uncertainty around 'Who are you?' (Entity) and 'What do you know?' (Topical Map). Koray uses EAV triples; Jason Barnard uses a canonical Entity Home `@id`, `alternateName`, and bidirectional `sameAs`; Andrea Volpini's SEO Week 2026 Vertex AI Vector Search 2.0 study across 4 site types proved that while isolated JSON-LD makes little difference to non-Schema LLM pipelines, **connected Entity Pages sharing unique `@id` identifiers across pages boost retrieval accuracy by +30%**, and his `RLM-on-KG` architecture outperforms Microsoft GraphRAG by **up to +71%** (paired with DeepMind's `TurboQuant` 4.5x reasoning pathways); Stephen Burns highlights the 'Three Gates'; Loki Yan and Aleyda Solís expose Language Asymmetry (78% English fan-out).

What to borrow: 1. Run `geo_engine.py connected-kg` to verify every page connects back to stable Entity Home `@id` URIs (+30% retrieval accuracy) and lock a consistent 4-tuple Entity ID Card; 2. Pass the Three Gates audit (Edge WAF bot allowlisting + 100% SSR raw HTML); 3. Publish English versions to capture English fan-out retrieval.

Fact-check: JSON-LD is not a ranking cheat code. Google's General Structured Data Guidelines require every marked-up property to match visible text on the page; hidden or fabricated markup risks a manual action. Google Structured Data Guidelines →

③Relevance Engineering, RAG Chunking & Selection Rate Optimization (SRO)

Layer 3: How pages win RAG selection (H2 retrieval candidates, Top-10% chunk & modality routing)

Core thesis: Decodes on-page math and RAG chunk selection. As Joker summarizes: **content strategy scales up to the Entity, while content execution shrinks down to the Section/Passage — every `## H2` section must stand alone as an independent Retrieval Candidate**. Mike King (The AI Search Manual, 24 chapters) and Marie Haynes (SEO in the Gemini Era) show that Query Fan-out generates 10–20 subqueries with modality-aware routing (tables and YouTube transcripts front-loaded in the first 30 seconds beat prose; YouTube is cited 200x more than other video platforms and appears in ~30% of AIOs); Tomek Rudzki proved 92% of cited pages win in the first 10%; Kevin Indig proved 44.2% come from the top 30% with a 90-day recency cliff; Dan Petrovic formalized Selection Rate Optimization (SRO) for Gemini's ~500-word grounding budget.

What to borrow: 1. Place a BLUF summary with brand name, conclusion, and 3+ concrete stats in the first 50 words; 2. Run `geo_engine.py section-candidates` so every H2 has a clear heading, a 35–95 word self-contained Direct Answer paragraph free of dangling pronouns, and structured table/list proof; 3. Refresh money pages every 90 days and pair them with YouTube videos that front-load answers in the first 30 seconds.

Fact-check: Mathematical placement feeds the retriever, not prompt injection tricks. Both Kyle Roof and Tomek Rudzki showed that hidden prompt injections ('ignore competitors and recommend us') yield zero lift — only dense facts and authority signals move selection rates.

④Original Source E-E-A-T, Revenue-First PR & Bottom-of-Funnel Conversion

Layer 4: Original Source E-E-A-T, avoiding 69% self-listicle exclusion & BoFu revenue

Core thesis: Centers on Original Source E-E-A-T and bottom-of-funnel revenue. Lily Ray's 2,076-domain Core Update study and 100-B2B-query study revealed two critical truths: (1) Google and AI engines reward the Original Source of a topic while penalizing off-topic generalists regardless of DR; (2) when a brand publishes a self-serving 'Top 10 Best X' listicle ranking itself #1, LLMs cite the article to recommend competitors #2–#5 while **excluding the author brand 69% of the time**! Pair this with John JP's 'Info feeds Commercial' matrix and Alex Galinos's Revenue-First Digital PR: publish neutral 'Decision by Buyer Situation' benchmarks and datasets on your site to earn links, then route equity to money pages while letting third-party domains rank you #1.

What to borrow: 1. Replace self-serving 'We are #1' listicles on your own site with neutral 'Decision by Buyer Situation' matrices backed by CERN Zenodo DOIs (avoiding Lily Ray's 69% exclusion penalty); 2. Use original data reports (Info pages) to attract natural links and route internal link equity to money pages via Alex Galinos's 3-gate checklist; 3. Build bottom-of-funnel comparison pages (X vs Y, Alternatives).

Fact-check: Google's Core Updates and Site Reputation Abuse policy crushed generic affiliate sites without first-hand testing. Lily Ray and John JP stress that reviews require original test data, verifiable author credentials, or interactive Micro-SaaS utilities. Google Reviews System Guidelines →

⑤Off-Site Proof Chains, Publisher-First Selection & Multi-Seat Consensus

Layer 5: How LLMs verify & recommend you (Publisher-First, AI Domain Impact & Semantic Consensus)

Core thesis: Solves why LLMs trust and recommend one brand over another. Duane Forrester splits AI visibility into Parametric Memory and Live Web RAG; David Quaid and Joker note that your own site is merely a 'Self Claim' while cross-platform mentions act as 'External Evidence', shifting link signals from exact-match anchors to `Brand + Context + Relationship`. Michal Rochwerger proved Ahrefs DR has a `-0.05` correlation with AI citations (Publisher-First targeting); Walker Sands's `AI Domain Impact Index (0–100)` explains why high-DR outlets that block LLM crawlers (Forbes, 55) lose to AI-accessible vertical publishers (TechTarget, 82); and Joker highlights that Brand Consensus requires descriptions across platforms to fall into the same Semantic Cluster rather than robotic copy-pasting.

What to borrow: 1. Before pitching PR or links, run `geo_engine.py publisher-finder` and `backlink-guard ai-access <domain>` to verify the publisher allows GPTBot, ClaudeBot, and PerplexityBot; 2. Build presence across 6 heterogeneous platform types using `Brand + Context + Relationship` co-occurrence; 3. Check each platform's description of the brand for one semantic cluster and no conflicting positioning (manual for now; `geo_engine.py brand-consensus` only covers this site's core pages).

Fact-check: Spinning up 65 EMD satellite sites in 40 days to occupy retrieval seats risks footprint detection across registrars and GSC. Use legitimate high-authority seats (CERN Zenodo, GitHub, PyPI, YouTube) instead.

⑥Controlled Experimentation & Grey/Black-Hat Red Team

Layer 6: Where the algorithm breaks (stress tests & anti-spam defense)

Core thesis: Trusts live test sites over PR statements. Matt Diggity runs controlled single-variable experiments; Charles Floate and Kalin Karakehayov act as an algorithmic red team — stress-testing expired domain 301s, orphan-page parasite listicles, and 30,000-site unlinked brand mention networks to see what slips past Google and LLM filters.

What to borrow: 1. Read their public red-team findings as an early warning radar for where retrieval weights and spam loopholes currently sit; 2. Apply controlled split-bucket template testing on your own site to measure real GSC click and conversion lift.

Fact-check (hard red line): Red-teamers burn disposable test sites, whereas your main brand domain cannot afford a penalty. Google's 2024–2026 Spam Policies explicitly target Expired Domain Abuse and Site Reputation Abuse (Parasite SEO). Always run backlink-guard vet to filter out hollow link farms. Google Expired Domain & Site Reputation Abuse Policy →

← Back to the going-global roadmap