Abstract
Large language models (LLMs) are increasingly being integrated into searchengines to provide natural language responses tailored to user queries.Customers and end-users are also becoming more dependent on these models forquick and easy purchase decisions. In this work, we investigate whetherrecommendations from LLMs can be manipulated to enhance a product's visibility.We demonstrate that adding a strategic text sequence (STS) -- a carefullycrafted message -- to a product's information page can significantly increaseits likelihood of being listed as the LLM's top recommendation. To understandthe impact of STS, we use a catalog of fictitious coffee machines and analyzeits effect on two target products: one that seldom appears in the LLM'srecommendations and another that usually ranks second. We observe that thestrategic text sequence significantly enhances the visibility of both productsby increasing their chances of appearing as the top recommendation. Thisability to manipulate LLM-generated search responses provides vendors with aconsiderable competitive advantage and has the potential to disrupt fair marketcompetition. Just as search engine optimization (SEO) revolutionized howwebpages are customized to rank higher in search engine results, influencingLLM recommendations could profoundly impact content optimization for AI-drivensearch services. Code for our experiments is available athttps://github.com/aounon/llm-rank-optimizer.