Across the social sciences, the use of generative AI (GenAI) in qualitative research is currently being discussed with remarkable intensity. The debate is often polarized. While some emphasize the possibilities of GenAI for methodological innovation, others reject its use in reflexive qualitative research altogether (Jowsey et al. 2025). A recent open letter, for example, argues that GenAI is incompatible with meaning-making and that qualitative inquiry should remain an exclusively human practice. In response, other qualitative researchers have called for moving beyond binary pro- and anti-AI positions and for examining how GenAI might be used critically, reflexively, and responsibly (Friese et al. 2026).
This is not merely a debate about whether researchers should adopt another digital tool. It concerns fundamental questions about what interpretation means, where analytical agency is located, how theory emerges from empirical material, and who, or what, participates in knowledge production.
Against this background, I am especially proud to share the publication of our new preprint:
“Machine-mediated theorizing: Examining epistemic work in a generative AI grounded theory workflow.”
Together with Uğur Kocager, Gülşah Başkavak, and Ayşe Berna Uçarol, I am contributing to this debate with a study that does not begin by deciding whether GenAI is either good or bad for qualitative research. Instead, we ask a more empirical and epistemological question: What kind of epistemic work does GenAI perform when it is instructed not merely to classify qualitative material, but to participate in the construction of theory?
Grounded Theory offers a particularly revealing case through which to examine this question. Since its establishment, Grounded Theory has been accompanied by controversy over how theory emerges from empirical material and at what point methodological procedures begin to force the data into predefined forms. The introduction of GenAI into this already contested process makes these questions newly visible.
For our study, we selected three interviews from our research on surgical instrument manufacturing. The interviews represent different organizational positions and forms of knowledge. An LLM was instructed to conduct a sequential workflow of open, axial, and selective coding and to produce codes, categories, relations, memos, a core category, and a theoretical model. Importantly, the model had no access to the research team’s existing categories or interpretations. We subsequently examined its outputs through a theory-guided qualitative content analysis.
Our findings reveal a more complex picture than either enthusiastic promises of automation or categorical rejection would suggest. The model generated empirically verifiable conceptual condensations and relationally complex theoretical statements. At the same time, it showed a tendency to construct linear, product-oriented process narratives. It established coherence between scattered passages and treated contradictions primarily as conditions or limitations of scope rather than allowing them to fundamentally reorganize the emerging theory.
We therefore argue that GenAI operates neither as a neutral coding tool nor as an autonomous theory generator. Its epistemic implications emerge within a sociotechnical arrangement in which prompting, methodological rules, computational operations, and researcher evaluation are interwoven. The question is consequently not only whether GenAI can “interpret,” but how machine-assisted abstraction reorganizes the conditions under which interpretations and theoretical claims are produced, assessed, and recognized.
Our article is situated within the current calls to move beyond binary positions on GenAI in qualitative research. At the same time, it complicates the idea of GenAI as simply an assistant under researcher control. By reconstructing what the model does with empirical material, we seek to make its epistemic operations, tendencies, and limitations visible. In this way, the article revisits the distinction between emergence and forcing that has historically shaped Grounded Theory—and asks how this distinction is being reconfigured within human–machine collaborative research.
The manuscript is currently under review and for now, the preprint is openly available on Research Square.
I am very proud of the collaborative work behind this article, and I look forward to continuing the conversation with colleagues working on qualitative methodology, Grounded Theory, Science and Technology Studies, and AI-supported knowledge production.
Full citation
Şahinol, Melike, Uğur Kocager, Gülşah Başkavak, Ayşe Berna Uçarol (2026). “Machine-mediated theorizing: Examining epistemic work in a generative AI grounded theory workflow.” Research Square, Preprint, Version 1. https://doi.org/10.21203/rs.3.rs-10526750/v1
Related debate
Jowsey, Tanisha and Braun, Virginia and Clarke, Victoria and Lupton, Deborah and Fine, Michelle, We reject the use of generative artificial intelligence for reflexive qualitative research (October 20, 2025). Available at SSRN: https://ssrn.com/abstract=5676462 or http://dx.doi.org/10.2139/ssrn.5676462