Imagine a chatbot that doesn't just respond to your questions but anticipates your needs and offers helpful recommendations. That's the promise of conversational recommender systems (CRS). But current chatbots, even those powered by large language models (LLMs) like ChatGPT, often struggle to provide truly helpful recommendations. Why? They lack two key ingredients: external knowledge and goal guidance. New research explores how to give LLMs these missing pieces, leading to more informative and proactive chatbot interactions. Researchers analyzed how LLMs perform in CRS tasks, finding that they excel in areas with ample built-in knowledge but falter when faced with unfamiliar topics. This makes it hard for them to offer relevant recommendations, especially in niche areas. The solution? Equip LLMs with access to external knowledge bases and guide them with clear conversational goals. This research introduces ChatCRS, a framework that does just that. ChatCRS breaks down complex conversations into smaller tasks, handled by specialized agents. A knowledge retrieval agent pulls relevant information from external databases, while a goal-planning agent sets the direction of the conversation. These agents work together, guided by a core LLM, to generate helpful responses and accurate recommendations. The results are impressive. ChatCRS significantly boosts the performance of LLMs in CRS tasks, leading to more informative and proactive conversations. It's a big step towards creating chatbots that can truly understand our needs and offer personalized recommendations. This research opens exciting new avenues for chatbot development. By combining the power of LLMs with external knowledge and goal guidance, we can create more helpful and engaging conversational experiences. The future of chatbots looks bright, with the potential to transform how we interact with technology and access information.