Empowering Social Work Research with Gpt: A Framework for Scalable, Ethical Thematic Analysis in Public Health Studies

Background and Purpose:

Thematic analysis is a cornerstone of qualitative inquiry in social work and public health, offering critical insights into the lived experiences of marginalized populations and informing responsive services. However, the process is labor-intensive, subjective, and resource-dependent—especially for practitioners in under-resourced community settings. These barriers can limit the capacity of frontline social workers and researchers to integrate qualitative evidence into timely practice and policy. This study proposes and evaluates a GPT-powered thematic analysis framework that supports rigorous, scalable, and ethically grounded analysis. By lowering the analytical burden, our approach enhances research-practice integration and democratizes access to qualitative evidence generation across diverse social work contexts.

Methods:

We applied the framework to 302 interview transcripts drawn from 11 peer-reviewed public health studies covering topics such as obesity prevention, healthcare communication, and food access. Using GPT-4 in a zero-shot, prompt-engineered setup, the model extracted themes at the chunk-, participant-, and study-levels through sequential prompt chaining. Each theme included a topic sentence, explanatory paragraph, and a supporting quote. Five independent evaluators (MPH/PhD-level) assessed GPT-generated themes using a rubric adapted from Braun and Clarke, rating them across four criteria: alignment with published themes, succinctness, explanation quality, and quote relevance. Intra-class correlation (ICC) was used to assess inter-rater reliability. Comparative t-tests evaluated mean performance against pre-defined thresholds.

Results:

GPT-generated themes achieved high concordance with researcher-identified themes, with a mean score of 3.05/4 across all criteria. All studies scored above threshold for explanation quality and succinctness, and 91% met standards for alignment. The ICC scores (0.87 overall; 0.92 for quote relevance) reflected strong consistency across raters. Thematic analysis was completed in under 30 minutes per study, with low implementation costs. Evaluators noted the model’s ability to reduce redundancy, maintain contextual fidelity, and support reproducibility—key concerns in qualitative social work research.

Conclusions and Implications:

This study presents a novel, low-cost, and time-efficient method to assist social work researchers and practitioners in conducting high-quality thematic analysis. The framework has strong implications for expanding research capacity among community-based organizations, integrating client voices into service design, and addressing barriers to evidence-based practice. While not a substitute for human judgment, GPT models can serve as collaborative tools to reduce analytical bias and increase transparency. By amplifying underrepresented perspectives and accelerating data-to-insight cycles, this approach strengthens the role of social work science in shaping equitable policies and practices.