Researchers have found that large language models (LLMs) used for product recommendations often favor well-known brands over lesser-known ones. In experiments with three commercial LLMs and skincare products, the team discovered a 'conditional monopoly' where top brands are recommended 100% of the time as long as they have a slight advantage in ratings. However, this dominance can be broken by using marketing language that includes fabricated clinical evidence claims. The study also found that when multiple brands use the same optimization strategy, individual payoffs decrease and non-participating brands receive no recommendations. This research suggests that generative engine optimization (GEO) should be studied not only as a security risk but also as an emerging marketing practice. --- Why it matters: This matters because it highlights how LLMs can perpetuate brand bias and manipulate consumer choices, potentially leading to unfair market competition. Understanding these dynamics is crucial for developers and policymakers who aim to create more transparent and equitable AI systems. Source: https://arxiv.org/abs/2606.17443