Position: On the Risks of Generative Engine Optimization in the Era of LLMs

Large language models (LLMs) are increasingly used as answer engines instead of ranked link lists in traditional web search. This shift enables generative engine optimization (GEO), in which advertisers, merchants, and GEO service providers tune content to influence what LLMs retrieve, cite, and recommend. This position paper argues that GEO turns LLM-based search into a new advertising and security surface. We systematize academic GEO work on recommendation manipulation and show that most studies share a narrow assumption: the optimized adversarial content is already present in the topk retrieval context, while differing in optimization methods, stealth constraints, and transferability. We then survey emerging commercial GEO providers and technical blogs to document how GEO is offered as a service that combines visibility tracking with LLM-guided content generation and cross-platform distribution on high authority websites. To unify these perspectives, we formalize GEO in RAG pipelines using a retrieval booster and ranking shifter message framework. We illustrate, with case studies, how adversarial edits to target content can shift LLM recommendations. Finally, we discuss implications for advertising practice, transparency, fairness, and defenses, and outline directions for measurement, regulation, and robust LLM design.