Abstract
Aim. This article explores the use of advanced prompt engineering in nursing research, with a focus on ChatGPT-generated Boolean search queries (BSQs).
Material and methods. The study compares the effectiveness of different models of ChatGPT: ChatGPT-3.5, ChatGPT-4.0, and ChatGPT-4omni, in generating high-quality BSQs for PUBMED. The prompting methods analysed involved Zero-Shot, Automated Chain-Of-Thought, Emotional Stimuli, Role-play, and Mixed-Methods prompting.
Results. The study found that ChatGPT-4omni, using Mixed-Methods prompting, achieved the highest quality scores, whereas ChatGPT-3.5, using zero-shot prompting, is the least effective. Significant variability in search outcomes was observed across different models and methods of prompting. The authors recommend ChatGPT-4omni as the most effective model for generating BSQs.
Conclusions. The study highlights the lack of standardized methods for prompt engineering in scientific research, complicating the use of large language models such as ChatGPT and underline the potential of ChatGPT to automate the preparation of systematic reviews and the development of search strategies. While ChatGPT proved valuable for generating search terms and synonyms, it often struggles to produce fully accurate BSQs. The article argues for the use of the latest ChatGPT models, along with advanced prompt engineering techniques, for scientific tasks. It also calls for further research to refine and standardise prompt engineering methods in nursing research.