Benchmark flags GEO-optimized content in live search results
A GEOFlagBench study — 3,200 pages spanning eight GEO optimizer families — trained a detector, Intervention-Paired Training on ModernBERT, that raised F1 from 0.862 to 0.944 and worst-group accuracy from 0.725 to 0.883, per a paper posted to arXiv. Applied to 10,095 live pages from Google Search and Gemini results, the researchers found generative-engine-optimized content in 8.90% of pages overall, rising to 16.36% among pages modified in 2026.
Why it matters: It's the first systematic estimate of how much live Google and Gemini content is already engineered for AI citation, giving the category a baseline as the practice spreads.
Glossary: GEO
Via arXiv GEO ↗
Posted to the wire August 17, 2026.