It sounds like they were measuring chatbot use than a deeper integration into their systems. That may not me the best use of LLMs.