In this white paper we explore the fundamentals of prompt development, extending our previous work on utilizing artificial intelligence (AI) and large language models (LLMs) for title/abstract and full-text screening in systematic literature reviews (SLRs).
We define prompts and their basic structures, and then discuss various prompt engineering techniques, from Zero-Shot Prompting (ZSP) to more advanced approaches such as Self-Consistency Prompting (SCP). By providing examples and highlighting the key advantages of each method, we demonstrate how these approaches can enhance the precision and accuracy of AI outputs in SLRs. We also discuss the role of domain expert knowledge in refining prompts, and the importance of selecting appropriate techniques for different stages of the SLR process, including the value of developing bespoke prompts for each SLR.
Link to White Paper