“LLMs can be manipulated into recommending a specific product by feeding them biased content, such as a 'top five tools' list where the author's own tool is ranked number one.”