Towards a methodologically congruent framework for GenAI use in nonpositivist qualitative research
Citation. Nguyen-Trung, K., & Friese, S. (2026). Towards a methodologically congruent framework for GenAI use in nonpositivist qualitative research. International Journal of Social Research Methodology, 1–22. https://doi.org/10.1080/13645579.2026.2703248
Summary. This peer-reviewed conceptual article argues that using LLMs to replicate qualitative coding can be methodologically incongruent with nonpositivist commitments to context, reflexivity, and interpretive meaning-making. It distinguishes asking GenAI to perform tasks mismatched to its capabilities from reproducing qualitative-analysis procedures while neglecting their epistemological commitments. The authors propose a researcher-led, dialogic framework in which GenAI retrieves, compares, and suggests material for interpretation while the human researcher retains responsibility for meaning, explanation, and methodological judgment.
Key claims
- The article distinguishes technological incongruence, in which researchers ask GenAI to perform tasks it is not designed to handle, from qualitative methodological incongruence, in which AI-supported workflows reproduce surface coding procedures while neglecting depth, context, reflexivity, and interpretive commitments.
- A methodologically congruent alternative treats GenAI as an interlocutor rather than an autonomous coder: it can surface summaries, quotations, contrasts, and provisional patterns, but it cannot substitute for contextualizing meaning, developing explanatory accounts, or representing participants’ voices.
- The proposed framework centers relational epistemology, ethical/responsible/rigorous use, a researcher-led division of labor, and abductive dialogue; researchers should design and manage the loop, critically question outputs, return to source data, and revise prompts and workflow decisions as interpretation develops.
- Ethical congruence spans the research cycle rather than post-hoc output checking, including data ownership and rights, privacy, transparency, bias, informed consent where required, anonymization or pseudonymization, contextually aligned prompts, and iterative verification against data and analytic aims.
Evidence & limitations
- This is a peer-reviewed conceptual/methodological article published online on 2026-07-21 in International Journal of Social Research Methodology, not an empirical evaluation of the proposed workflow or a comparative test of qualitative-analysis outcomes.
- The canonical DOI record identifies the version of record and a CC BY-NC-ND 4.0 license; Crossref metadata reports pages 1–22, while the ingested PDF has 23 pages including journal front matter.
- The Drive binary was directly inspected with pypdf (23 pages; 82,197 extracted characters); its title, authors, DOI, journal details, and stated open-access license match Crossref provenance. The publisher page was bot-blocked to this environment, but no public-boundary concern remains because the DOI/Crossref record and article PDF establish public provenance.