Dental Patient Simulation
Synthetic patient twins built from real Google reviews. 9-practice validation in Plano, TX.
Setup
9 dental practices in Plano, TX stratified into 3 tiers by Google rating x log(review count). Reviews scraped via Outscraper. One synthetic "twin" built per practice from its last 12 months of patient reviews.
Method
Each twin ranked the other 8 practices based on their patient blurbs (excluding self). The aggregate ranking was compared against ground truth. A negatives-only variant was also tested using only 1-3 star reviews.
Results
Spearman rank correlation: 0.82. All 3 high-tier practices ranked above all 3 low-tier practices. The negatives-only variant produced even sharper results with fewer tokens. Total cost: ~$5 via Outscraper.
Limitations
Review-writing patients skew younger and more digitally engaged. Geographic concentration in one metro limits generalizability. Phase 2 plans MRP post-stratification using Census ACS data.