Show or Tell? Modeling the evolution of request-making in Human-LLM conversations