Rethinking survey development in health research with AI-driven methodologies

Abstract

Artificial intelligence (AI), particularly large language models (LLMs), offers new opportunities to address methodological challenges in survey development for health research. Traditional approaches, such as manual item generation, cognitive interviewing, and post-hoc psychometric validation, are time- and resource-consuming, and vulnerable to undetected issues that emerge only after large-scale data collection. These limitations, which appear in the early stages, can spread to later phases, leading to costly revisions and weakened construct validity. This paper introduces a conceptual framework for integrating AI-driven techniques throughout the survey development cycles. Drawing on natural language processing, automated text analysis, real-time data monitoring, and predictive modeling, the framework outlines how AI tools can help researchers proactively uncover linguistic nuances, identify hidden patterns, and refine instruments with greater speed and rigor, ultimately enhancing validity, inclusivity, and interpretive richness. Rather than replacing existing practices, these tools are positioned as a complementary support that, when used responsibly and contextually, can enhance methodological rigor, improve efficiency, and reduce respondent burden. The paper also emphasizes ethical considerations, including transparency, interpretability, and mitigation of bias. By combining AI's computational power with human expertise and critical reflexivity, this approach aims to foster more responsive, inclusive, and valid instruments for health-related research and interventions.