Competitor-aware GEO strategy selection beats single-heuristic baselines

A paper posted to arXiv on August 27 formalizes generative-engine-optimization strategy selection as a competitor-aware problem: as adoption of content-rewriting strategies grows, the optimal strategy for any single page changes, and single-strategy methods can fall below an unoptimized baseline once competition is high. The authors’ two-phase method — searching the strategy space with Bayesian Optimization of Combinatorial Structures, then fine-tuning a language model on the results to recommend strategy combinations — beat the existing AgenticGEO baseline by more than 18% on a synthetic competitive benchmark and by 8% to 21% on out-of-distribution test sets, while degrading only 11.4% of its own score across the full range of simulated competitor adoption.