Monte Carlo Simulations for Generative Optimization
Daniel Wang
Founder · UC Berkeley MIDS
TL;DR
Sill applies Monte Carlo simulations — a technique from finance and physics — to map how AI models decide which brands to recommend. Small changes like price positioning or adding "AI-powered" can shift citation rates by 40% or more.
Monte Carlo simulations have been used for decades in finance, physics, and engineering to model uncertainty. Now, we're applying the same principles to understand AI search behavior.
How It Works
Monte Carlo simulations define key variables, generate thousands of controlled prompt variations, test across multiple LLMs, and analyze citation patterns.
Monte Carlo simulations for GEO define key variables, generate thousands of controlled prompt variations, test them across multiple LLMs, and analyze citation patterns.
- Define Variables: : We identify the key factors that influence AI recommendations (price point, feature set, positioning, etc.)
- Generate Variations: : We create thousands of prompt contexts with controlled variable changes
- Run Simulations: : Each variation is tested across multiple LLM providers
- Analyze Results: : We identify patterns and thresholds in citation behavior
A Real Example
A CRM startup saw citations drop from 15% to 2% above $100/user/month, while adding "AI-powered" to positioning increased citations by 40%.
Simulations can reveal that small changes — like price positioning or adding "AI-powered" to product descriptions — shift AI citation rates by 40% or more.
Consider a CRM startup competing with Salesforce and HubSpot. Traditional SEO would focus on keywords like "best CRM software." But LLMs don't process queries the same way.
Our simulations revealed that: - At price points below $50/user/month, the startup was cited 15% of the time - Above $100/user/month, citations dropped to 2% - Adding "AI-powered" to the positioning increased citations by 40%
This data is actionable. It tells you exactly what to adjust and how much.
Beyond Vanity Metrics
Decision Boundary data is precise and actionable — unlike impressions or CTR, you know exactly where you stand and what to change to move the needle.
Decision Boundary data provides precise, actionable measurements of AI visibility — unlike impressions or click-through rates, you know exactly what to change.
Unlike impressions or click-through rates, Decision Boundary data is precise. You know exactly where you stand and what it takes to move the needle.
That's the power of Generative Optimization.