A Pragmatic Approach to AI in Qualitative Research
2026
We have a new article in the Annual Review of Sociology on artificial intelligence (AI), computational social science (CSS), and qualitative research, particularly qualitative data analysis.
Our argument is straightforward: AI is not something social scientists can simply “opt in” to or “opt out” of. AI already shapes the tools we use to search, transcribe, analyze, proofread, and circulate research. It also affects high-stakes domains where social scientists conduct research and help shape practice, including health, organizational life, and social inequality.
We make the case for a pragmatic approach to research that neither fetishizes technologies nor rejects them a priori. Instead, we ask what each tool makes possible, what it may distort, and whether it serves our methodological principles and ethical commitments.
Drawing on decades of work connecting in-depth qualitative data (including fieldnotes, interviews, and other texts) with computational social science, we identify five patterns of engagement likely to become more prominent with AI: streamlining workflows, scaling projects, hybrid analysis combining breadth and depth, studying computation as an object of analysis, and principled non-use.
For us (and others), the promise of advanced computation is not simply speed or efficiency, but greater accuracy and new analytic possibilities. In our own work, small local language models paired with human review can help detect errors, check coding, map broader patterns, and identify disconfirming cases without transmitting data to external servers. We discuss these possibilities alongside their risks and potential mitigations.
W. E. B. Du Bois’s data visualizations and C. Wright Mills’s filing cabinet offer historical examples of social scientists pragmatically repurposing technologies with complex histories in pursuit of rigorous insight. We share examples from scaled, team-based ethnographies of inequality, aging, and life with terminal illness. These include applications in a recent American Sociological Review article on terminal cancer and agency at the end of life and a Social Science & Medicine article on life with dementia. Across these projects, we have used computational methods to extend rather than replace human insight. We hope that discussing our process and decisions is useful as others decide what makes sense for their own work.
Because the technical side of AI changes (very) rapidly, we focus on methods, workflows, and tensions that persist even as tools change: the role of researchers working alongside communities, pressures to keep up with the “algo-Joneses,” and the distinction between scientism and rigor. The article and supplement link to open-source code, resources, and workflows we have been compiling for several years, with generous feedback from colleagues and students.
Article:
Abramson, Corey M., Tara Prendergast, Zhuofan Li, and Daniel Dohan. 2026. “Qualitative Research in an Era of Artificial Intelligence: A Pragmatic Approach to Data Analysis, Workflow, and Computation.” Annual Review of Sociology 52(1):35–61.
https://doi.org/10.1146/annurev-soc-011824-104836
Supplemental appendix:
https://www.annualreviews.org/content/journals/10.1146/annurev-soc-011824-104836#supplementary_data
Open-source code, resources, and workflows:
https://github.com/Computational-Ethnography-Lab/teaching