JMIR Publications

ABSTRACT

Background:

The landscape of online health information-seeking is shifting from traditional search engines to include conversational interactions with Large Language Models (LLMs). Qualitative studies on user goals and strategies for using LLMs for health-related purposes are lacking.

Objective:

To identify user goals, interaction patterns, and prompting strategies for health-related LLM interactions using a cross-cultural longitudinal approach.

Methods:

We conducted a 14-day longitudinal diary study with 36 participants (18 in the United States and 18 in India). A total of 277 diary entries recording health-related LLM use were analyzed using the Goals Associated with Health Information Seeking (GAINS) framework, descriptive statistics, and inductive thematic analysis. Our participants used a variety of LLMs including Chat GPT, Google Gemini, Meta AI, and Claude

Results:

Participants use LLMs for a wide range of health purposes including searching for general health information, symptom diagnosis, analyzing health information, and fitness tracking. Across these varied cases, participants reported high levels of satisfaction using LLMs for health-related purposes (96% US, 97% India). They report high levels of success across GAINS goals including action planning, reassurance, understanding health problems and hope (>70% of entries rated as completely or mostly successful). Participants in the United States were more likely to include personal context in prompts (71/132, 54% US vs 42/144, 29% India; P<.001) and use expressive prompts (29/131, 22% US vs 11/137, 8% IN). Participants in India were more likely to use directive prompts (40/137, 29% IN vs 12/131, 9% US). Indian participants reported a significantly higher rate of positive emotional impact ("felt better" 94% vs 64% US), while US participants were more likely to feel "the same" (34% vs 3% India).

Conclusions:

User strategies for online health information seeking are shifting from navigational browsing to conversational prompting. Cultural contexts moderate how users interact with and feel using LLMs for health. These differences appear more in prompting styles than in user goals and motivations. Understanding these nuances is critical for designing globally relevant, safe, and effective health AI systems.