Can Generative Artificial Intelligence Interpret Culture? Assessing ChatGPT's Role in Qualitative Health Research

The rapid integration of artificial intelligence into research has raised urgent questions about whether generative models can contribute meaningfully to qualitative inquiry, particularly in health contexts where cultural nuance and contextual interpretation are essential. This study evaluates ChatGPT's capacity to support ethnographic analysis by comparing AI-assisted outputs with a previously conducted human-led analysis of 20 interviews with college students managing Type 1 diabetes. Guided by Spradley's four-phase ethnographic interview method, it examines ChatGPT-4o's ability to identify cultural domains, construct taxonomies, articulate contrastive dimensions, and synthesize higher-order themes. While ChatGPT efficiently generated surface-level summaries and facilitated early data organization, its analytic contributions remained largely descriptive. The model struggled to identify symbolic tensions, contextual meanings, and culturally embedded interpretations that human analysts derived through iterative reading, reflexivity, and familiarity with the lived experience of chronic illness. AI-generated taxonomies lacked internal structure, and thematic interpretations failed to account for the institutional, relational, and identity-based dynamics shaping students' experiences. These findings suggest that although generative AI can serve as a useful assistive tool for preliminary coding and pattern recognition, it cannot replicate the interpretive depth central to qualitative health research. The study highlights the methodological implications of incorporating AI into qualitative inquiry, underscoring the need for critical oversight and researcher reflexivity when integrating AI-assisted approaches. As the use of AI expands in health education and behavior research, understanding its limitations is essential for ensuring that culturally grounded, meaning-rich insights remain at the center of qualitative analysis.