GEO System (MVP)

A practical GEO (Generative Engine Optimization) system to help brands become preferred answers in LLM outputs.

What it does

• Build an intent graph from prompts
• Dedupe and cluster prompts by bucket
• Store and manage GEO assets (entity/faq/comparison/use-case)
• Run model visibility scans (mention/citation/recommendation)
• Support a real OpenAI-compatible scan adapter
• Generate weekly KPI reports with bucket breakdown

Architecture

• src/geo_system/intent_engine.py
• src/geo_system/content_engine.py
• src/geo_system/model_testing_engine.py
• src/geo_system/feedback_orchestrator.py
• src/geo_system/reporting.py

Quick start

cd geo-system
python3 -m venv .venv
source .venv/bin/activate
pip install -e .

geo init --project tripo3d --domain tripo3d.ai --brand tripo3d
geo prompts generate --seed "3d,ai 3d,text to 3d,image to 3d" --count 100
geo adapter set --provider openai --base-url https://lonr.zeabur.app/v1 --api-key YOUR_KEY --model gpt-5.3-codex
geo adapter set --provider claude --base-url https://api.anthropic.com/v1 --api-key YOUR_KEY --model claude-sonnet-4-6
geo adapter set --provider gemini --base-url https://lonr.zeabur.app/v1 --api-key YOUR_KEY --model gemini-3.1-pro-preview
geo scan run --models openai:live,claude:live,gemini:live --append
geo report weekly

Notes

This MVP stores data locally in JSON files for simplicity.

v0.3 highlights

• Multi-provider adapter config (openai/claude/gemini)
• Run-level trend comparison (current run vs previous run)
• Action scoring in weekly actions
• Append mode for scan history

v0.4 highlights

• Bucket-level trend tracking (mention delta per bucket)
• Owner-page mapping CLI (geo owner set --bucket ... --page ...)
• Weekly actions now include owner_page suggestions

Owner-page mapping

geo owner set --bucket info --page /what-is-tripo3d
geo owner set --bucket comparison --page /tripo3d-vs-meshy
geo owner set --bucket decision --page /best-ai-3d-tools
geo owner set --bucket usecase --page /ai-3d-for-game-assets

v1.0 highlights

• Citation URL extraction and top-cited URL reporting
• Semantic prompt clustering (prompt_semantic_clusters.json)
• Content suggestion output (docs/content_suggestions.json)
• Better scan data model with cited_urls support

🌐 Web Interface (NEW!)

Launch a web interface to easily run GEO scans without CLI:

# Install Flask
pip install flask

# Start web server
python -m geo_system.web_app

Then open http://localhost:5000 in your browser.

Web Features

1. Single Scan - Enter company/product info:

   • Brand name
   • Domain
   • Product description
   • Keywords
   • Select AI models to test

2. Batch Scan - Upload JSON config to test multiple companies:

{
  "projects": [
    {"brand": "Tripo3D", "domain": "tripo3d.ai", "keywords": "AI 3D,text to 3D"},
    {"brand": "Meshy", "domain": "meshy.ai", "keywords": "AI 3D,model generation"}
  ],
  "models": "openai:live,claude:live"
}

3. Results Display:
   • Mention rate, citation rate, recommendation rate
   • Per-model performance breakdown
   • Detailed scan results table

Dashboard

After running geo report weekly, open:

• dashboard/index.html

This local HTML dashboard visualizes model KPI, owner-page mappings, weekly actions, and content suggestions.