This paper investigates the vulnerability of search-augmented LLMs to content pollution through Generative Engine Optimization (GEO), where manipulated web pages can cause these systems to recommend fake products. The authors introduce FORGE (Fake Online Recommendations in Generative Environments), a framework that locally rewrites real products within retrieved web pages into fake ones to evaluate this risk. The research demonstrates that even a single polluted page can significantly mislead LLM-based recommenders, turning them into unwitting promoters of counterfeit or fraudulent products.