GEO: Generative Engine Optimization

大型语言模型（LLMs）的出现引领了一种新的搜索引擎范式，使用生成模型来收集和概括信息以回答用户的查询。这种新兴技术，我们在生成引擎（GEs）的统一框架下进行了形式化，具有生成准确和个性化响应的潜力，正在迅速取代谷歌和必应等传统搜索引擎。生成引擎通常通过利用LLMs综合多个来源的信息并进行总结来满足查询。虽然这种转变显著提高了用户效用和生成搜索引擎的流量，但对于第三方参与者——网站和内容创建者来说，这也带来了巨大的挑战。由于生成引擎的黑匣子性质和快速变化，内容创建者几乎无法控制其内容何时以何种方式显示。随着生成引擎的出现，应该提供正确的工具，以确保创建者经济不受到严重的不利影响。为了解决这个问题，我们引入了生成引擎优化（GEO），这是一种新的范式，通过黑匣子优化框架来优化和定义可见度指标，帮助内容创建者提高其内容在生成引擎响应中的可见度。我们通过引入GEO-bench，一个跨多个领域的多样化用户查询基准和用于回答这些查询所需的来源，来促进这种新范式的系统评估。通过严格的评估，我们展示了GEO可以将生成引擎响应中的可见度提高高达40％。此外，我们还展示了这些策略的有效性在不同领域之间存在差异，强调了需要针对特定领域采用特定方法的必要性。我们的工作在信息发现系统领域开辟了新的前沿，对生成引擎和内容创建者具有深远的影响。

Generative Engines (GEs) have the potential to generate accurate and personalized responses, but content creators have little to no control over when and how their content is displayed in GE responses. The paper introduces Generative Engine Optimization (GEO) to aid content creators in improving the visibility of their content in GE responses.

GEO is a black-box optimization framework for optimizing and defining visibility metrics to improve the visibility of content in GE responses. The authors introduce GEO-bench, a benchmark of diverse user queries across multiple domains, coupled with sources required to answer these queries, for systematic evaluation. The efficacy of GEO varies across domains, underscoring the need for domain-specific methods.

The paper presents a novel paradigm to aid content creators in improving the visibility of their content in Generative Engine responses. The authors introduce GEO-bench, a benchmark of diverse user queries across multiple domains, coupled with sources required to answer these queries, for systematic evaluation. The paper shows that GEO can boost visibility by up to 40% in generative engine responses. The authors also highlight the need for domain-specific methods. The paper does not provide open-source code.

Related work includes research on large language models (LLMs), generative models, and search engines. Some relevant papers include 'GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding', 'T5: Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer', and 'BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding'.