How It Works

Architecture overview

User Input (URL or file)
    ↓
┌─────────────────────────────────────────┐
│         E-GEO Pipeline                  │
│  ┌─────────┐  ┌────────┐  ┌─────────┐  │
│  │Analyzer │→ │ Ranker │→ │Rewriter │  │
│  └─────────┘  └────────┘  └────┬────┘  │
│                                ↓        │
│                          ┌──────────┐  │
│                          │ Indexer  │  │
│                          └──────────┘  │
└─────────────────────────────────────────┘
    ↓
geo-output/ (report, optimized content, schema)

The same four agents run either in-process through the Python runtime behind the egeo CLI, or host-executed through Claude Code /geo slash commands. A small runtime adapter layer (egeo/runtimes.py) keeps the two paths sharing one source of truth.

The 4 agents

1. Analyzer

Extracts content, scores it against the 10 universal GEO features, identifies gaps, and writes analysis.json:

{
  "url": "https://example.com",
  "geo_score": 67,
  "features": {
    "ranking_emphasis": 5,
    "user_intent": 9,
    "competitive_diff": 4
  },
  "gaps": ["missing social proof", "weak competitive positioning"]
}

2. Ranker

Simulates how an AI search engine would rank the content against competitors, predicts positions, and explains why competitors rank higher. Per the E-GEO paper (arXiv:2511.20867), competitive framing produces the strongest immediate ranking improvement.

3. Rewriter

Rewrites content applying the 10 GEO features while preserving brand voice and factual accuracy — no fabricated statistics, testimonials, or ratings.

#	Feature	What it means
1	Ranking Emphasis	Frame as best/top choice
2	User Intent Alignment	Directly address search intent
3	Competitive Differentiation	Highlight unique advantages
4	Social Proof	Reviews, testimonials, stats
5	Compelling Narrative	Persuasive, engaging language
6	Authoritativeness	Expert, confident tone
7	Unique Selling Points	Clear differentiators
8	Urgency Signals	Scarcity or time-sensitivity
9	Scannable Format	Headings, bullets, structure
10	Factual Accuracy	Never fabricate data

4. Indexer

Generates JSON-LD schema markup (SoftwareApplication, Organization, Article, Product, Service, FAQPage), meta tags, and implementation checklists.

The research behind E-GEO

E-GEO is based on the paper “E-GEO: Optimizing Content for Generative Engines” (arXiv:2511.20867), building on the foundational Princeton GEO study (Aggarwal et al., KDD 2024).

Key findings the toolkit operationalizes:

Competitive framing produces the strongest immediate ranking lift among tested GEO strategies.
A universal strategy (all 10 features together) outperforms individual heuristics.
The 10 GEO features consistently appear in higher-ranking content across ChatGPT, Perplexity, and Gemini.

See the paper for full methodology; results vary by content quality and competition.

Validation layer

When run through Claude Code, E-GEO validates outputs against ground truth using MCP servers:

MCP server	Purpose	Criticality
Brave Search	Competitor analysis, SERP data, ground truth	High
Chrome DevTools	Rendered DOM validation, performance metrics	High
fetch	Simple text scraping (fallback)	Medium

When MCP servers are unavailable, E-GEO still runs but marks outputs “Low Confidence”. See MCP Server for setup.

Auto-triggered skills

Skill	Trigger	Purpose
competitive-analysis	Questions about competitors	Analyze the competitive landscape
content-scoring	Requests to score content	Evaluate against the 10 GEO features
schema-generator	Structured data needs	Generate JSON-LD markup
validation-doctor	MCP setup issues	Verify dependencies and configuration
geo-loop	`/geo:loop <domain>`	Enforce the loop-mode run contract