Abstract
The advent of large language models (LLMs) has ushered in a new paradigm ofsearch engines that use generative models to gather and summarize informationto answer user queries. This emerging technology, which we formalize under theunified framework of generative engines (GEs), can generate accurate andpersonalized responses, rapidly replacing traditional search engines likeGoogle and Bing. Generative Engines typically satisfy queries by synthesizinginformation from multiple sources and summarizing them using LLMs. While thisshift significantly improves $\textit{user}$ utility and $\textit{generativesearch engine}$ traffic, it poses a huge challenge for the third stakeholder --website and content creators. Given the black-box and fast-moving nature ofgenerative engines, content creators have little to no control over$\textit{when}$ and $\textit{how}$ their content is displayed. With generativeengines here to stay, we must ensure the creator economy is not disadvantaged.To address this, we introduce Generative Engine Optimization (GEO), the firstnovel paradigm to aid content creators in improving their content visibility ingenerative engine responses through a flexible black-box optimization frameworkfor optimizing and defining visibility metrics. We facilitate systematicevaluation by introducing GEO-bench, a large-scale benchmark of diverse userqueries across multiple domains, along with relevant web sources to answerthese queries. Through rigorous evaluation, we demonstrate that GEO can boostvisibility by up to $40\%$ in generative engine responses. Moreover, we showthe efficacy of these strategies varies across domains, underscoring the needfor domain-specific optimization methods. Our work opens a new frontier ininformation discovery systems, with profound implications for both developersof generative engines and content creators.