Prompt engineering in ChatGPT for literature review: practical guide exemplified with studies on white phosphors

Recent advancements in large language models (LLMs) such as ChatGPT have been transforming the ways we approach science tasks, including data analysis, experimental design, writing, and literature review. However, due to the lack of specialized knowledge and inherent issues such as plagiarism and hallucinations (i.e., false or misleading outputs), it is necessary for users to verify the output information. To address these issues, prompt engineering has become a significant task. In this study, we evaluate the performance of different prompt styles for extracting information from literature abstracts and emphasize the importance of prompt engineering for such scientific tasks. The literature on white phosphor materials is used for this study due to the availability of important and quantitative information in the abstracts. Through detailed comparative and quantitative evaluation, we provide guidance on preparing suitable and effective prompts based on the types of information sought.