GEO Optimizer

Generative Engine Optimization (GEO) for AI search visibility.

What is GEO?

GEO is NOT traditional SEO. AI search engines (ChatGPT, Perplexity, Claude, Gemini) work fundamentally differently:

They fetch raw HTML - Many don't render JavaScript
They have size limits - Crawlers abandon pages >1MB HTML
They evaluate confidence - Hedged language ("maybe", "possibly") ranks 3x lower than confident assertions
They need machine-readable metadata - AgentFacts/NANDA protocol for AI agent discovery

Quick Audit Workflow

1. Technical Visibility Check

# Run full GEO audit
python3 ${CLAUDE_SKILL_DIR}/scripts/audit-geo.py https://example.com

# Check HTML size and JS dependency
curl -sI https://example.com | grep -i content-length

2. Content Authority Analysis

# Check hedge density of your content
python3 ${CLAUDE_SKILL_DIR}/scripts/check-hedge-density.py --url https://example.com/blog/post

# Or analyze text directly
python3 ${CLAUDE_SKILL_DIR}/scripts/check-hedge-density.py --text "Your content here"

3. Agent Infrastructure Setup

# Generate AgentFacts schema for your domain
python3 ${CLAUDE_SKILL_DIR}/scripts/generate-agentfacts.py \
  --domain example.com \
  --agent-name "Example Service" \
  --capabilities text,image \
  --output /.well-known/agent-facts

Core Metrics

Hedge Density Score

Measures uncertainty language in content. Target: <0.2%

Hedge words to avoid:

maybe, possibly, perhaps, might, could be
however, although, it seems, arguably
some believe, it appears, potentially

Why it matters: Research shows confident citations rank 3x higher in AI responses. Every hedge word signals uncertainty to LLMs.

See: references/hedge-density.md

HTML Size Budget

Maximum: 1MB raw HTML. Above this threshold:

18% of pages abandoned by crawlers
Content truncation risks
Slower indexing

Check with: curl -sI URL | grep content-length

See: references/technical-visibility.md

JS Dependency Risk

AI crawlers vary in JavaScript rendering:

Rendering crawlers: GPTBot, PerplexityBot (sometimes)
Non-rendering: ClaudeBot, many others

If your content requires JS to display, you're invisible to ~40% of AI crawlers.

Discovery Gap

Startups face a "recency wall" - 30:1 visibility disadvantage vs established domains:

Site Age	Visibility	Strategy
<2 years	~3.3%	Web-augmented: Reddit, referring domains
2+ years	~99%	GEO content: entity density, hedge reduction

See: references/discovery-strategies.md

AgentFacts Setup

The NANDA protocol provides machine-readable metadata for AI agent discovery. Place at /.well-known/agent-facts:

{
  "@context": "https://nanda.dev/ns/agent-facts/v1",
  "id": "nanda:example.com",
  "agent_name": "urn:agent:example:com",
  "endpoints": {
    "static": ["https://api.example.com/v1/agent"]
  },
  "capabilities": {
    "modalities": ["text"],
    "authentication": {
      "methods": ["oauth2", "jwt"]
    }
  },
  "trust": {
    "certification": "self-attested",
    "human_oversight": "true"
  }
}

See: references/agentfacts-schema.md

Audit Report Sections

A complete GEO audit covers:

Technical Visibility
- HTML payload size (target: <1MB)
- JS dependency assessment
- Crawler accessibility
Content Authority
- Hedge density score (target: <0.2%)
- Confidence assessment
- Entity density
Discovery Strategy
- Site age analysis
- Recency wall detection
- Recommended approach
Agent Infrastructure
- AgentFacts presence
- Schema validation
- NANDA compliance

Scripts

Script	Purpose
`audit-geo.py`	Full GEO audit for a URL
`check-hedge-density.py`	Analyze content confidence
`generate-agentfacts.py`	Create NANDA-compliant schema

Reference Documentation

File	Contents
`hedge-density.md`	Confidence scoring methodology
`agentfacts-schema.md`	NANDA protocol specification
`technical-visibility.md`	Crawler behavior & limits
`discovery-strategies.md`	Startup vs incumbent approaches

Example Commands

# Full audit with report
python3 ${CLAUDE_SKILL_DIR}/scripts/audit-geo.py https://mysite.com --output report.md

# Quick hedge check
python3 ${CLAUDE_SKILL_DIR}/scripts/check-hedge-density.py --url https://mysite.com/about

# Generate AgentFacts for SaaS
python3 ${CLAUDE_SKILL_DIR}/scripts/generate-agentfacts.py \
  --domain myapp.com \
  --agent-name "MyApp AI" \
  --capabilities text \
  --auth oauth2,apikey

Key Takeaways

GEO != SEO - Different crawlers, different rules
Confidence wins - Hedge density <0.2% is the target
Size matters - Stay under 1MB HTML
JS is risky - Content must work without rendering
Age affects strategy - Startups need web-augmented signals
AgentFacts is future-proofing - Implement NANDA now

geo-optimizer