This paper investigates the effectiveness of using ChatGPT as a top-n conversational recommendation system. The authors build a rigorous pipeline to simulate realistic user interactions with ChatGPT, where users provide initial instructions and then reprompt with feedback to refine recommendations. The study explores the impact of popularity bias in ChatGPT's recommendations and compares its performance with baseline models. Key findings include: 1. **Reprompting with Feedback**: Reprompting ChatGPT with user feedback significantly improves recommendation relevancy. 2. **Popularity Bias**: ChatGPT exhibits popularity bias, but this can be mitigated through prompt engineering. 3. **Performance Comparison**: ChatGPT outperforms random and traditional recommender systems in terms of recommendation quality. The paper also discusses the limitations of the current system, such as the need for large text datasets and the use of relatively old movie releases. Future work could involve comparing ChatGPT with other large language models and state-of-the-art recommendation algorithms.This paper investigates the effectiveness of using ChatGPT as a top-n conversational recommendation system. The authors build a rigorous pipeline to simulate realistic user interactions with ChatGPT, where users provide initial instructions and then reprompt with feedback to refine recommendations. The study explores the impact of popularity bias in ChatGPT's recommendations and compares its performance with baseline models. Key findings include: 1. **Reprompting with Feedback**: Reprompting ChatGPT with user feedback significantly improves recommendation relevancy. 2. **Popularity Bias**: ChatGPT exhibits popularity bias, but this can be mitigated through prompt engineering. 3. **Performance Comparison**: ChatGPT outperforms random and traditional recommender systems in terms of recommendation quality. The paper also discusses the limitations of the current system, such as the need for large text datasets and the use of relatively old movie releases. Future work could involve comparing ChatGPT with other large language models and state-of-the-art recommendation algorithms.