Empirical Benchmark of Generative Engine Optimization (GEO): Citation Share and Hallucination Rates in Advanced LLMs (2026)irical Benchmark of Generative Engine Optimization (GEO): Citation Share and Hallucination Rates in Advanced LLMs (2026)

This dataset and empirical study introduces a systematic framework for evaluating Generative Engine Optimization (GEO). We benchmarked top-tier LLMs (GPT-4o, Claude 3.5 Sonnet, Google Gemini 1.5 Pro, and Perplexity Pro) using a standardized dataset of B2B SaaS queries. The primary metrics evaluated include Citation Share (the frequency an AI directly links to a primary source) and Contextual Hallucination Rates. Our findings demonstrate the impact of Agent Presence Optimization (APO) elements like llms-full.txt and agent.json on improving model fidelity and retrieval accuracy. Conducted by KusiAI Labs.s