This paper introduces a formal model for designing and analyzing interactive platforms where Large Language Models (LLMs) offer sponsored clarifying follow-up prompts to ambiguous user queries. It investigates the trade-offs between end-to-end joint optimization and decoupled modular mechanisms, demonstrating that the VCG mechanism can achieve efficient and truthful outcomes in a unified system, while modular approaches suffer from unbounded strategic inefficiency.