Wheeler (2026) — Technological Reflexivity in Practice: How MAXQDA, NVivo, and ChatGPT Shape Qualitative Survey Analysis

TL;DR: Wheeler introduces “technological reflexivity” — the practice of critically examining how digital tools co-produce analytic decisions, not just the data they process. Drawing on a reflexive comparison of three tools (MAXQDA, NVivo, ChatGPT) on 1,300+ climate survey responses, she shows that different software architectures shape not just efficiency but what gets noticed, what gets written, and what counts as a finding. Reflexivity is distributed, not individual.

Problem

Qualitative researchers have developed sophisticated practices of reflexivity about their own positionality, assumptions, and interpretive choices. The parallel literature on CAQDAS — qualitative data analysis software — has long warned that tools shape analytic outcomes, not just automate them. Yet the standard practice in methods reporting is to name the tool used without explaining how it shaped the analysis.

The arrival of generative AI in CAQDAS platforms intensifies this gap. ChatGPT is not just a faster version of NVivo’s code-and-retrieve function. Its conversational interface, its opacity about how interpretations emerge, and its tendency to generate plausible-sounding syntheses create new methodological risks that are not captured by existing reflexivity frameworks.

Wheeler’s problem is both empirical and conceptual. Empirical: at the time of writing, no systematic comparisons existed between GenAI tools embedded within CAQDAS platforms (like MAXQDA’s AI Assist) and public LLMs (like ChatGPT). Conceptual: existing frameworks treat reflexivity as an individual researcher practice, but if tools are co-producing meaning, reflexivity must be distributed across humans, tools, and institutional infrastructures.

Approach

Wheeler analyzed the same corpus with all three tools and systematically documented how each shaped her interpretive decisions. The data: 1,300+ qualitative survey responses from young people (aged 8–25) imagining the life of a fictional peer “Alex” in 2050 (climate theme), generating over 5,000 short responses. This emotionally charged, youth-generated corpus requires interpretive sensitivity to idiom, cultural context, and affective register — not just descriptive pattern recognition.

The analysis is organized around Paulus and Lester’s (2023) four guiding questions for technological reflexivity:

1. How do tools shape methodological choices?
2. How do tools shape what counts as knowledge?
3. How do tools affect researcher-participant relationships?
4. How does tool design constrain or enable analysis?

Wheeler’s positionality is explicitly flagged: she is a professional MAXQDA trainer with long experience in the platform, a user of NVivo through her institution’s subscription, and a relative newcomer to ChatGPT for research purposes. This asymmetry shapes her reading of each tool’s affordances and limitations.

The paper follows Reflexive Thematic Analysis (Braun & Clarke 2022) as its analytic framework, and applies it to the meta-level question of how tools mediated the analysis itself.

AI’s Role

AI appears in three distinct roles in this paper — as a tool being compared, as an object of methodological critique, and as a medium through which reflexivity is made visible.

As a comparison tool, ChatGPT is positioned as extending existing CAQDAS logics in two directions: increased speed and conversational fluency, but reduced transparency about how interpretations are generated. Wheeler finds that ChatGPT’s outputs were harder to trace back to specific data than MAXQDA’s code-retrieve results — not because the outputs were wrong, but because the interface encourages accepting synthesis without seeing the underlying segments.

This is a specific operationalization of epistemic-flattening: not that AI produces wrong themes, but that its conversational interface makes it easier to accept themes without interrogating them.

Epistemological Stance

Social constructionist with posthumanist elements. Wheeler explicitly frames CAQDAS and GenAI as “infrastructures that co-produce meaning” rather than neutral tools. This is a posthumanist epistemological position: interpretive responsibility is shared across human and non-human actors, including the software architecture, institutional subscriptions, and training data embedded in each platform.

The distributed reflexivity concept draws on Barad’s (2003) notion of “intra-action” — meaning does not precede the interaction between researcher and tool but emerges through it. This is philosophically ambitious and sets Wheeler’s paper apart from most CAQDAS literature, which remains within a more conventional researcher-tool distinction.

The analysis does not aspire to reliability or replicability in any conventional sense. Its goal is interpretive depth and methodological transparency — explicitly Big Q by the standards of brailas-ai-qualitative-research-2025.

Rigor and Trustworthiness

Rigor is established through methodological self-disclosure: Wheeler documents her relationship to each tool, her analytic moves with each, and the moments where tool design led her to different interpretive choices. The audit trail is the researcher’s reflexive account, not a reliability metric.

The paper’s reflexivity about its own limitations is unusual. Wheeler acknowledges that her greater fluency with MAXQDA may have produced a favorable reading of its affordances. She flags the absence of systematic comparisons as a gap her study begins to fill but cannot close.

The study’s trustworthiness claim rests on coherence and transparency rather than replication. This is appropriate for its epistemological commitments, but researchers expecting generalizable benchmarks will find it insufficient.

Limitations

The single-researcher design — Wheeler alone using all three tools — means the comparison reflects one researcher’s strategies, skills, and preferences. A multi-researcher comparison would be methodologically stronger, though it would also dilute the reflexive depth that makes this paper valuable.

The youth climate dataset has specific characteristics (short responses, fictional scenario, UK-charity administration) that limit generalizability to other qualitative contexts. Whether the tool-mediation effects Wheeler documents appear in interview or focus-group data is an open question.

The paper does not provide systematic evidence that different tools produced different findings — it documents that different tools shaped different analytic strategies. The link from strategy differences to outcome differences is argued but not demonstrated.

The institutional-political dimensions of tool access (who can afford which software, whose institution subscribes to what) are raised as ethical concerns but not developed into an equity analysis. dahal-genai-qualitative-nepal-2024 develops this dimension more fully in a Global South context.

Connections

• llm-qualitative-research — the broader landscape; this paper extends reflexivity debates into the GenAI era
• prompt-engineering — Wheeler argues prompts are methodological choices equivalent to analytic memos; they must be documented as part of the audit trail
• ai-research-ethics — algorithmic opacity as an ethical issue; if you can’t trace how an interpretation emerged, you can’t adequately account for it
• epistemic-flattening — Wheeler operationalizes the concept in a tool-comparison context: the risk is accepting synthesis without interrogation
• brailas-ai-qualitative-research-2025 — aligned posthumanist/constructionist position; Brailas develops the theoretical argument, Wheeler demonstrates it empirically
• xu-ai-thematic-analysis-2026 — parallel reflexive TA with ChatGPT; compare the two approaches to documenting AI-mediated analysis
• paulus-marone-qdas-discourse-2024 — parallel study examining how QDAS software shapes discourse analytic practice
• validity-trustworthiness — distributed reflexivity as an approach to trustworthiness in tool-mediated qualitative research