Applying Large Language Models to Sponsored Search Advertising

With the increasing availability of powerful large language models (LLMs), the generation of textual marketing content has become more accessible. In this research, we examine the potential to tailor an LLM for application to search engine advertising (SEA). That is, we develop and evaluate an “application layer” that sits on top of an open-source LLM to generate ad text “fine-tuned” to the SEA context. With a goal of maximizing clicks to improve online visibility in a cost per click (CPC) setup, we experimentally test our framework in two empirical settings. Our results demonstrate the superior performance of a human-in-the-loop generative artificial intelligence (AI) approach to advertising content generation compared with ads created by humans and standard LLMs. We show that our approach yields improved performance, but potentially incurs a higher CPC, making it necessary to balance content optimization and cost. Our research demonstrates the performance gains achievable through the development of tailored LLM-based applications. Using our framework, we also identify boundary conditions that appear to limit the benefits of using generative AI in support of SEA, offering substantive insights to both practitioners and researchers.