IF-GEO: Conflict-Aware Instruction Fusion for Multi-Query Generative Engine Optimization

TL;DR

IF-GEO tackles the challenge of making a single document optimally visible across multiple heterogeneous queries within a limited content budget. It introduces a diverge-then-converge workflow that first discovers latent queries and per-query edit preferences, then fuses them into a Global Revision Blueprint to guide blueprint-guided edits, all while optimizing for cross-query stability using metrics $WCP$, $DR$, and $WTR$. The approach is validated on multi-query GEO benchmarks, showing superior overall visibility and reduced variance across queries compared to baselines such as static GEO heuristics, RAID, and Auto-GEO. The work demonstrates that explicit cross-query coordination via a structured blueprint yields more robust GEO performance, with practical implications for generating source-visible answers in Generative Search Engines.

Abstract

As Generative Engines revolutionize information retrieval by synthesizing direct answers from retrieved sources, ensuring source visibility becomes a significant challenge. Improving it through targeted content revisions is a practical strategy termed Generative Engine Optimization (GEO). However, optimizing a document for diverse queries presents a constrained optimization challenge where heterogeneous queries often impose conflicting and competing revision requirements under a limited content budget. To address this challenge, we propose IF-GEO, a "diverge-then-converge" framework comprising two phases: (i) mining distinct optimization preferences from representative latent queries; (ii) synthesizing a Global Revision Blueprint for guided editing by coordinating preferences via conflict-aware instruction fusion. To explicitly quantify IF-GEO's objective of cross-query stability, we introduce risk-aware stability metrics. Experiments on multi-query benchmarks demonstrate that IF-GEO achieves substantial performance gains while maintaining robustness across diverse retrieval scenarios.

Figures (3)

• Figure 1: Challenges of GEO. Revision requests of different queries can be conflicting and competitive under a limited content budget. GEO have no idea which query to follow.
• Figure 2: Overview of the IF-GEO methodology. IF-GEO follows a "diverge-then-converge" paradigm. In Phase I, the system mines a representative query set and elicits query-specific editing requests, which may be redundant or conflicting. In Phase II, IF-GEO performs conflict-aware instruction fusion to synthesize a unified global revision blueprint, which guides controlled editing to produce an optimized document with stable visibility across queries.
• Figure 3: Effect of query expansion size $N$ on performance and stability. (a) Overall objective performance (Mean). (b) Win--tie rate (WTR). (c) Worst-case performance (WCP). (d) Volatility metrics: variance (VAR) and downside risk (DR; lower is better).