Automated Thematic Analysis of Health Content

Prompt

The full prompts for this paper are provided in Appendix A1 of the supplementary materials available at JAMIA Open online. The paper describes an iterative prompt development process:

• Initial zero-shot experiments were conducted, followed by single-shot and multi-shot prompting strategies.
• Prompts were refined over multiple rounds, adjusting phrasing, task instructions, and exemplar formatting to align model predictions with expert-coded labels and maximize F1-score.
• Representative examples of each theme were incorporated as few-shot demonstrations.

Usage Notes

This prompt is from the paper “Automating inductive thematic analyses of health content using large language models” (Hairston et al., 2025).

• Task: Automating the traditionally manual process of inductive thematic analysis on social media health data.
• Model: GPT-4.
• Input: Social media posts on health topics.
• Approach: Iterative prompt refinement from zero-shot to multi-shot with representative theme examples.
• Key finding: LLMs can produce thematic analyses comparable to human researchers for health-related social media content.