Source
urlhttps://doi.org/10.1177/16094069251337583
rawraw/chatzichristos-2025-qualitative-research-in-the-era-of-ai-a-return-to-positivism-or-a-new-paradigm.pdf

TL;DR: A TAM survey combined with semi-structured interviews with European qualitative researchers reveals a generational divide: early-career researchers embrace AI’s efficiency and scale, experienced researchers worry about reflexivity and interpretive depth. Chatzichristos raises the concern that widespread AI adoption among junior researchers could gradually pull the field back toward positivism — a “positivism creep” effect operating not through explicit argument but through methodological norm formation.

Problem

Qualitative research emerged in the mid-20th century as a principled alternative to the positivist assumptions that governed quantitative social science. Its core methodological commitments — contextual understanding, interpretive depth, reflexivity, the researcher as the primary instrument — were developed in conscious contrast to the assumption that social phenomena could be measured, aggregated, and generalized.

AI’s adoption in qualitative research creates pressure on these commitments. Automated, data-driven approaches to theme identification and coding are structurally aligned with positivist assumptions: categories emerge from the data (or appear to), patterns are discovered rather than constructed, and consistency (replicability) is the quality criterion. The concern Chatzichristos raises is not that individual researchers are making positivist arguments — it is that AI adoption reshapes methodological defaults, particularly for researchers still forming their identities and practices.

The historical trajectory matters here: Grounded Theory, initially developed within a positivist framework by Glaser and Strauss (1967), was later reconstructed by Charmaz (2006) as a constructivist methodology. AI may be reversing that trajectory — not through theoretical argument, but through the practical affordances of the tools researchers use.

Approach

The study employs the Technology Acceptance Model (TAM) — a framework from information systems research that measures perceived usefulness and perceived ease of use as predictors of technology adoption. This is combined with semi-structured interviews targeting European qualitative researchers.

The TAM design is unusual in the qualitative methods literature, which rarely applies quantitative survey instruments to study its own practitioners. The justification: TAM provides systematic measurement of adoption attitudes across a population, allowing comparisons across career stages. The interviews add interpretive depth to the survey findings.

The sampling strategy specifically targeted European researchers, with the explicit goal of surfacing structural inequalities in AI adoption — access to resources, geographic variation, and gender — that US-centric studies may miss.

AI’s Role

AI appears in this paper primarily as an object of sociological study — the phenomenon being investigated is how qualitative researchers perceive and adopt AI, not how AI can be used to do better qualitative research. The paper’s relationship to AI is diagnostic and cautionary.

The concern is structural: AI’s automated features produce outputs that look like qualitative findings but are generated through processes (statistical pattern recognition, token prediction) that are epistemologically incompatible with interpretivism’s assumptions about knowledge production. Researchers who adopt AI without understanding this incompatibility may not recognize the shift in what their analysis is actually doing.

Epistemological Stance

Interpretive-sociological, with an explicit historical grounding in the development of qualitative research traditions. The paper situates the AI question within the longstanding positivism/interpretivism debate in social science, and its concern about “positivism creep” is a contribution to that debate rather than a purely methodological argument.

The use of TAM is an interesting internal tension: TAM was developed within a positivist information systems tradition and applies the assumption that adoption attitudes can be measured and aggregated. Using it to study qualitative researchers’ attitudes toward AI involves some methodological irony that the paper does not address.

Rigor and Trustworthiness

The mixed-method design — TAM survey plus semi-structured interviews — provides both systematic measurement and interpretive depth. The combination is well-suited to the research questions: TAM captures the scope and direction of the generational divide; interviews capture the reasoning and concerns behind it.

The sample is explicitly European, with a stated rationale. The limitation this creates — generalizability to other contexts — is acknowledged and addressed through the call for future research on structural inequalities.

The argument about positivism creep is the paper’s most important claim, and it is the least empirically secured. The evidence for it is the generational divide in attitudes: junior researchers embrace AI, experienced researchers worry. The inference from this finding to a trajectory of positivism creep is a theoretical claim, not a demonstrated outcome.

Limitations

The study measures attitudes toward AI in qualitative research, not actual practice. How researchers perceive AI’s usefulness is not the same as how they actually use it, and there is an established gap in the technology adoption literature between stated attitudes and behavior.

The “positivism creep” concern is compelling but speculative. Younger researchers may form different attitudes as they gain experience, or the field may develop norms and training that counteract the drift. The paper acknowledges this uncertainty but does not model the conditions under which positivism creep would or would not occur.

The paper does not engage deeply with the counterargument: that AI may enable new forms of interpretive research that were previously impossible at scale, rather than simply replicating positivist patterns. friese-caai-framework-2026 and costa-abductivai-2025 represent this alternative possibility; Chatzichristos does not address them.

The structural inequalities the paper identifies — geography, gender, resource access — are named as future research directions without being investigated. dahal-genai-qualitative-nepal-2024 begins to address the geographic dimension from a Global South perspective.

Connections

What links here