Counter-GEO-Bench: Evaluating Defenses Against Information-Distorting Generative Engine Optimization

2026-09-02

AI Summary

研究提出Counter-GEO-Bench，用于评估防御生成引擎优化导致的误导信息问题，通过对比实验发现C-GEO Guard方法能有效降低攻击成功率。

Abstract

Generative engine optimization (GEO) enables content producers to increase the visibility of their web pages in generative search engines, but the same techniques can deliver targeted misinformation when adversaries publish ordinary-looking GEO-optimized documents that victim large language models (LLMs) retrieve and synthesize into distorted answers. No existing benchmark evaluates defenses against this threat under controlled conditions. Therefore, we present Counter-GEO-Bench, a defense benchmark that pairs 247 human-verified, quality-gated queries with information-preserving and information-distorting GEO rewrites, and evaluates defenses on attack success rate (ASR), false positive rate, and answer quality across three victim LLMs. Under Counter-GEO-Bench, three off-the-shelf defenses (Granite Guardian, Llama Guard 3, and NeMo Self-Check Fact-Checking) reduce ASR by at most 5.7% relative, while Granite Guardian's reduction is not statistically significant. Safety-taxonomy guardrails target policy violations, while GEO misinformation passes through them as fluent informational content. To this end, a lightweight benchmark baseline, C-GEO Guard, is proposed, reducing ASR by 47.6% relative with near-zero utility loss, which proves threat tractable.

Problem

• Generative Engine Optimization
• Misinformation
• Large Language Models
• Defense Evaluation

Innovation

• Counter-GEO-Bench
• Generative Engine Optimization (GEO)
• Misinformation Defense
• Large Language Models (LLMs)
• C-GEO Guard