As the aviation industry embraces advanced training technologies, large language models (LLMs) offer new opportunities for enhancing flightcrew training through adaptive, personalized, and interactive learning experiences. However, the utility of LLMs depends heavily on prompt engineering. Prompt engineering involves designing inputs that guide the model toward generating expected outputs of sufficient accuracy and relevance. However, the integration of human factors (HF) principles may help to ensure that LLMs align with the cognitive, psychological, and operational realities of flightcrew training environments. In this article, we evaluate the outcomes of unstructured versus structured prompts for an instructional design task, finding that structured, context-rich, and HF-aligned prompts often produce more actionable and operationally relevant training materials for certain aspects of flightcrew training applications. Through a comparative analysis of structured and unstructured prompts using two prominent LLMs (ChatGPT-4o and DeepSeek), we evaluated how output quality may be impacted by prompting techniques, including the PERFECT framework. Our findings demonstrate how structured, HF-informed prompting may improve scenario fidelity and support training for procedural skills and monitoring behaviors. This article presents a set of evidence-based practices for integrating generative AI (GenAI) into training design to better reflect the realities of modern flightcrew operations. Future research should explore human-in-the-loop validation to further bridge the gap between AI capabilities and specific aviation training requirements.
Duruaku et al. (Thu,) studied this question.