This work revisits the role of prompt engineering in personalized recommendation tasks. The paper comprehensively evaluates various LLM-based prompting strategies for user preference modeling and shows that their effectiveness depends heavily on data context and personalization granularity.