kevin-liu-01 SEO/GEO Optimization
GEO = Generative Engine Optimization. AI search engines don't rank pages — they cite sources. Being cited is the new "ranking #1."
Read wiki/tools/seo-geo.md for the full playbook if available.
The 9 Princeton GEO Methods
Ranked by measured visibility boost:
| Method | Boost | How to Apply |
|---|---|---|
| Cite Sources | +40% | Add authoritative citations and references |
| Statistics Addition | +37% | Include specific numbers and data points |
| Quotation Addition | +30% | Add expert quotes with attribution |
| Authoritative Tone | +25% | Use confident, expert language |
| Easy-to-understand | +20% | Simplify complex concepts |
| Technical Terms | +18% | Include domain-specific terminology |
| Unique Words | +15% | Increase vocabulary diversity |
| Fluency Optimization | +15-30% | Improve readability and flow |
| -10% | AVOID — hurts visibility |
Best combination: Fluency + Statistics = maximum boost.
Workflow
Step 1: Audit Current State
# Check meta tags and schema
curl -sL "https://example.com" | rg -i "<title>|<meta name=\"description\"|<meta property=\"og:|application/ld\+json" | head -20
# Check AI bot access
curl -s "https://example.com/robots.txt"
# Check sitemap
curl -s "https://example.com/sitemap.xml" | head -50
Verify AI bot access in robots.txt:
- Googlebot, Bingbot (required)
- GPTBot, ChatGPT-User (OpenAI/ChatGPT)
- ClaudeBot, anthropic-ai (Claude)
- PerplexityBot (Perplexity)
Blocking Google-Extended is not the same as blocking Googlebot. If the goal is AI visibility, blanket AI-bot blocking is self-sabotage.
Step 2: Entity Engineering
The 5 Required Rules (from Kevin's playbook):
- Keyword placement — primary keyword in at least 4 of 6: URL slug, title tag, meta/social title, H1, at least one H2, body copy intro
- Entity placement — primary entity + supporting entities in title, H1, intro, core sections, schema
- Entity gap analysis — extract entities from top competitors, add missing relevant entities. Do not add irrelevant brands.
- JSON-LD schema — place in page
<head>, not only through client-side injection - Schema entity reinforcement — mirror on-page entities inside schema fields (
name,description,about,sameAs,knowsAbout,mentions,author,publisher)
Step 3: Schema Markup
Baseline graph: Organization/Person + WebSite + page-specific type + BreadcrumbList.
Use sameAs to disambiguate across Wikipedia, Wikidata, LinkedIn, YouTube, GitHub, Crunchbase. Prefer one @graph block with @id references. Use absolute URLs and ISO dates.
FAQPage Schema (+40% AI visibility):
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "What is [topic]?",
"acceptedAnswer": {
"@type": "Answer",
"text": "According to [source], [answer with statistics]."
}
}]
}
Meta Tags Template:
<title>{Primary Keyword} - {Brand} | {Secondary Keyword}</title>
<meta name="description" content="{Compelling description with keyword, 150-160 chars}">
<!-- Open Graph -->
<meta property="og:title" content="{Title}">
<meta property="og:description" content="{Description}">
<meta property="og:url" content="{Canonical URL}">
<meta property="og:image" content="{Absolute image URL 1200x630}">
<meta property="og:image:width" content="1200">
<meta property="og:image:height" content="630">
<meta property="og:image:type" content="image/png">
<meta property="og:image:alt" content="{Image alt text}">
<meta property="og:type" content="website">
<meta property="og:site_name" content="{Brand}">
<meta property="og:locale" content="en_US">
<!-- Twitter Cards (namespace is still twitter:, not x:) -->
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:site" content="@{handle}">
<meta name="twitter:creator" content="@{handle}">
<meta name="twitter:title" content="{Title}">
<meta name="twitter:description" content="{Description}">
<meta name="twitter:image" content="{Absolute image URL}">
<meta name="twitter:image:alt" content="{Image alt text}">
Robots directive for large social cards:
<meta name="robots" content="max-image-preview:large">
Or in Next.js metadata:
robots: {
index: true,
follow: true,
googleBot: {
index: true,
follow: true,
"max-image-preview": "large",
"max-snippet": -1,
"max-video-preview": -1,
},
},
Step 4: AI Citability Engineering
Models cite passages that are direct, self-contained, fact-rich, structurally legible, and make sense outside context.
5-part citability model:
- Answer block quality — question-based H2, direct answer in first 1-2 sentences, named entities, precise claims
- Self-containment — names subject explicitly, no dangling pronouns, at least one concrete fact
- Structural readability — one H1, clean H2-H3 nesting, short paragraphs, lists for steps, tables for comparisons
- Statistical density — percentages, dollar values, dated claims, named studies, benchmarks
- Uniqueness — original data, case studies, experiments, screenshots
Target: 134-167 word answer blocks.
Step 5: Content Structure
- Use "answer-first" format (direct answer at top of each section)
- Clear H1 > H2 > H3 hierarchy
- Bullet points and numbered lists
- Tables for comparison data
- Short paragraphs (2-3 sentences max)
- Question-led H2s for AI Overview extraction
Step 6: llms.txt
Machine-readable summary at domain root:
- What the site is, what sections matter, canonical pages, core facts
- 10-30 most important pages with clear descriptions
- Aligned with real site structure
Not a replacement for good structure, schema, or crawlability.
Platform-Specific Strategies
ChatGPT
- Branded domain authority — cited 11% more than third-party
- Content freshness — update within 30 days (3.2x more citations)
- Backlinks — >350K referring domains = 8.4 avg citations
- Bing index matters; entity consistency across site/schema/profiles
Perplexity
- Allow PerplexityBot in robots.txt
- Use FAQ Schema (higher citation rate)
- Host PDF documents (prioritized for citation)
- Community validation, Reddit footprint, sourced quotable content
Google AI Overview
- Optimize for E-E-A-T
- Structured data (Schema markup), topical authority (content clusters + internal linking)
- Authoritative citations (+132% visibility)
- Question-led H2s, direct answer after heading, tables, FAQ, visible
dateModified
Microsoft Copilot / Bing
- Bing indexing required for citation
- Microsoft ecosystem signals (LinkedIn, GitHub mentions help)
- Page speed < 2 seconds, clear entity definitions
Claude
- Brave Search indexing (Claude uses Brave, not Google)
- High factual density (data-rich content preferred)
- Clear structural clarity (easy to extract)
Competitive Comparison Capture
Capture branded commercial-intent queries directly:
- Publish exact-match
vs,review, andalternativespages - Per competitor: one standalone review page + one comparison page
- Structure: question-led H2s, quotable verdict sentences, side-by-side tables, FAQ blocks
- Firsthand testing required — sign up, use products, take notes and screenshots
GEO Scorecard
| Category | Weight |
|---|---|
| AI citability and extractability | 25% |
| Brand authority and off-site signals | 20% |
| Content quality and E-E-A-T | 20% |
| Technical foundations | 15% |
| Structured data | 10% |
| Platform optimization | 10% |
Bands: 90-100 elite, 75-89 strong, 60-74 mixed, 40-59 weak, 0-39 critical.
Validation
# Schema validation
open "https://search.google.com/test/rich-results?url=https://example.com"
open "https://validator.schema.org/?url=https://example.com"
# Indexing check
open "https://www.google.com/search?q=site:example.com"
open "https://www.bing.com/search?q=site:example.com"
Next.js OG Implementation
buildPageMetadata pattern
When 3+ pages repeat the same openGraph/twitter block, extract a helper that
returns { openGraph, twitter, alternates } from a single input. Every
public page should call this instead of hand-writing the block. See the
og-metadata-audit skill for the full pattern.
Social bot middleware
For maximum reliability, intercept social bots in middleware.ts and return
a minimal HTML document with just meta tags. This avoids bots hitting React
loading shells, Clerk auth walls, or heavy JS bundles.
The middleware file must be named middleware.ts and the export must
be named middleware or default. Next.js silently ignores proxy.ts or
other names.
htmlLimitedBots
Complement the middleware with htmlLimitedBots in next.config.ts to serve
lightweight HTML to SEO/AI crawlers the middleware doesn't catch:
htmlLimitedBots: /Googlebot|Bingbot|GPTBot|ClaudeBot|PerplexityBot|.../,
Canonical inheritance trap
In Next.js App Router, if a child page doesn't export its own
alternates.canonical, it inherits the parent layout's — usually /. Every
public page needs its own canonical URL. buildPageMetadata fixes this by
including alternates: { canonical } in every call.
metadataBase
Must resolve to the production domain. If using env vars, verify
NEXT_PUBLIC_SITE_URL is set in Vercel. Fallback chains
(VERCEL_PROJECT_PRODUCTION_URL → VERCEL_URL) resolve to *.vercel.app,
causing Twitter to cache images against the wrong origin.
Related Skills
seo-audit— comprehensive technical SEO auditingog-metadata-audit— OpenGraph/Twitter card audit and DRY patternscontent-strategy— content planning and publishing rhythmsocial-draft— platform-optimized content drafting