Assembling the evidence…
GenAI is increasingly used in qualitative research, yet much current practice remains methodologically incongruent. Attempts to make LLMs replicate qualitative coding often create a mismatch between underlined qualitative values (e.g. depth, context, and reflexivity) and actual AI-supported analytic practices (e.g. surface-level extraction and automated coding). This article distinguishes between technological incongruence, where GenAI is asked to perform tasks it is not designed to handle, and qualitative methodological incongruence, where AI workflows reproduce the surface procedures of qualitative analysis while neglecting its interpretive and reflexive commitments. We propose a shift from ‘AI as coder’ to ‘AI as interlocutor’. We outline key tenets for methodologically congruent GenAI use in non-positivist qualitative research: relational epistemology, ethical, responsible and rigorous use, researcher’s lead role, and abductive dialoguing. GenAI can support qualitative analysis by surfacing, retrieving, comparing, and suggesting material for interpretation, but meaning-making remains the responsibility of the human researcher.
We propose a shift from ‘AI as coder’ to ‘AI as interlocutor’.
coherent conceptual proposal grounded in critique, without empirical validation
This article distinguishes between technological incongruence... and qualitative methodological incongruence
the distinction is defined and illustrated across Sections 2.1 and 2.2
Attempts to make LLMs replicate qualitative coding often create a mismatch between... qualitative values... and actual AI-supported analytic practices
multiple examples support the direction, though representativeness is not systematically established
GenAI can support qualitative analysis by surfacing, retrieving, comparing, and suggesting material for interpretation, but meaning-making remains the responsibility of the human researcher.
consistent with the researcher-led framework and ethical workflow discussion
Coding is a human crutch – a scaffold built to support our limited processing power.
provocative conceptual framing that is argued plausibly but not empirically demonstrated
Derived from the full evaluation — not a separate score.
Strengths
The distinction between technological incongruence and qualitative methodological incongruence gives the debate a structured vocabulary. It helps separate tool-capability mismatches from deeper conflicts between automated workflows and nonpositivist qualitative values.
↳ Abstract; §§2.1–2.2
Table 1 grounds the critique in a detailed deconstruction of a one-turn thematic-analysis prompt. This makes the methodological argument more concrete than a purely abstract rejection of AI-assisted coding.
↳ Table 1; §2.1
Rather than simply rejecting GenAI, the article proposes researcher-led, dialogic use through relational epistemology, ethical practice, researcher leadership, and abductive dialoguing. This gives the paper a visible pathway into methodological discussion and tool design.
↳ §§4.1–4.4
Limitations
The article makes field-level claims about the replication approach and LLM suitability from a narrative review without transparent search or inclusion procedures. This leaves uncertainty about how representative the cited failures and examples are.
↳ Introduction; §§2.1–3
The four tenets are useful as principles, but they do not yet provide detailed decision rules, validation evidence, or clear failure-case handling for practitioners. The article itself describes them as value-based principles rather than prescriptions.
↳ §4
The disclosure identifies a commercial connection to QInsights.ai, while the article discusses commercial qualitative AI platforms and advocates AI-supported qualitative workflows. The disclosure is transparent, but the relationship is relevant to how readers interpret the framework endorsements.
↳ Introduction; end matter / disclosure statement
The scores are driven by the article’s strong conceptual organization and its concrete analysis of prompt-level failures in Table 1. The contribution is meaningful because it gives the field a vocabulary for separating technological limitations from methodological misalignment, and it proposes a constructive alternative to outright rejection. The main constraints are the narrative, non-systematic evidence base and the fact that the proposed tenets remain broad principles rather than a tested workflow. The disclosed commercial connection to QInsights.ai is handled transparently, but it remains relevant because the article discusses commercial AI-supported QDA tools and author-adjacent frameworks.
Sound3.6/5.0
The score reflects a genuine conceptual contribution: the article distinguishes technological from qualitative methodological incongruence and advances the “AI as interlocutor” framing. The contribution is mainly synthetic and normative rather than a new empirical or formally validated framework.
We propose a shift from ‘AI as coder’ to ‘AI as interlocutor’.
The argument proceeds in a clear sequence from defining incongruence, to diagnosing examples, to proposing broad tenets for congruent GenAI use. Methodological Rigour is limited by reliance on illustrative examples, limited operational decision rules, and less critical scrutiny of the preferred CAAI, NITA, and QBA approaches.
These tenets are not doctors’ prescription; they are value-based principles
The article is clearly organized around memorable distinctions and labels that make the reasoning easy to follow. Some phrasing about what LLMs or chatbots cannot do is more categorical than the implementation-dependent technical issue warrants.
technological incongruence... qualitative methodological incongruence
The article engages current debates over rejection versus responsible integration of GenAI and connects them to Big Q, small q, and medium Q traditions. Contextual positioning is moderated by the narrative review format, partial treatment of more successful GenAI-coding evidence, and limited explicit discussion of author-adjacent frameworks.
a parallel position has emerged in support of the responsible and critical integration
Lower confidence on Positioning — domain match limited.
Watch-outs4 of 4 checks
The conceptual argument is broadly coherent, but the paper sometimes moves from selected narrative examples to broad statements about current practice and LLM capability. These are watch-outs about claim proportionality rather than contradictions that undermine the paper’s thesis.
The paper transparently discloses a commercial conflict of interest and GenAI proofreading use. The disclosed COI is relevant because the article discusses commercial AI-supported QDA tools and endorses frameworks adjacent to the authors’ work.
Flags: 1 declared / 5 total
64 references verified via PubMed, Crossref, or OpenAlex.
No retraction notice found in Retraction Watch.
Sources: Retraction Watch ✓PubPeer (coming soon)
Medium3.4/5.0
The article addresses an active problem for qualitative researchers deciding whether and how to use GenAI in nonpositivist analysis. It identifies relevant users and decision contexts, but does not specify adoption by particular institutions or standards bodies.
the question of how LLMs fit meaningfully into qualitative research is rarely addressed
The paper provides usable principles, examples, and workflow concepts such as the human-designed-and-managed loop. Readiness is limited because the framework is not tested and the tenets are not converted into validated implementation criteria.
AI outputs should not be accepted at face value.
The framework is principle-based and could apply across many nonpositivist qualitative traditions and AI tool environments. Generalizability is bounded because the article explicitly excludes or problematizes small-q, positivist, and coding-reliability uses.
non-positivist qualitative research
The paper is positioned within a fast-moving debate, including recent rejection and integration movements and adjacent frameworks such as CAAI, NITA, and QBA. Its trajectory is plausible but depends on later operationalisation, empirical testing, and independent uptake.
a growing body of work reporting use cases for GenAI in qualitative research
Checked against Unpaywall on 25 Jul 2026.
Nabu’s assessment, alongside the field’s view.
Evaluation Metadata
AI-powered, human-governed. Something look off? Contact us to request a review.