Abstract
<title>Abstract</title> <p>Discussions on generative AI (GenAI) in knowledge-intensive fields continue to be shaped by polarized narratives that oscillate between promises and dangers. These narratives rarely address the context-dependent and historically contested epistemic practices of qualitative research, into which GenAI is entering. Grounded Theory (GT) provides a revealing case for examining this problem, since its establishment has been accompanied by a long-standing controversy over how theory emerges from empirical material and when methodological procedure turns into forcing. This article examines how GenAI performs epistemic work when it is instructed not only to classify interview material but to transform it into codes, categories, relations, a core category, and a theoretical model. The study draws on three interviews from a research project on surgical instrument manufacturing, as they represent different organizational positions and forms of knowledge. An LLM was instructed to conduct a sequential workflow of open, axial, and selective coding without access to the research team’s existing categories or interpretations. The results generated by GenAI (e.g., codes, memos, paradigm models, etc.) were analyzed using a theory-guided qualitative content analysis. It became apparent that the model generated empirically verifiable conceptual condensations and relationally complex theoretical statements. It further favored linear, product-oriented process narratives, established coherence between scattered passages, and considered contradictory material more as a condition or limitation of scope rather than allowing it to fundamentally reorganize the emerging theory. GenAI does not operate as a neutral coding tool nor as an autonomous theory generator. This is because its epistemic implications are embedded in a sociotechnical context in which, e.g., prompting, methodological rules, computational operations, and researcher evaluation are interwoven within a human-machine collaborative network. By reconstructing these processes, we can understand how machine-assisted abstraction redefines the distinction between emergence and forcing that has historically shaped GT.</p>