The Machine at the Research Table: AI, Reflexivity, and the Future of Qualitative Inquiry

AI and Qualitative Research: From Research Tool to Methodological Challenge

Artificial intelligence is no longer a polite “future guest” waiting outside the research office; in this article, it has already walked into healthcare, education, and scientific inquiry. Dellafiore, Saba, Collaro, and Artioli open by making the background unmistakably clear: AI, as a field spanning computer science, engineering, mathematics, linguistics, and psychology, is rapidly reshaping research methods; predictive AI uses machine learning and statistical models to identify patterns and make decisions, while generative AI produces text, images, audio, and video through large foundational models (Dellafiore et al., 2026). Yet the article does not sing a simple hymn to technology. Instead, it places the controversy in full view: when AI enters qualitative research, promises of efficiency and convenience immediately collide with epistemological, ethical, and deeply human questions about whether the researcher can still remain at the center of the work.

To bring those tensions down from abstraction into lived research experience, the authors used a qualitative interpretative design built on semi-structured interviews and reflexive thematic analysis. The participants were not casual users but 14 expert qualitative researchers working in Italian universities and healthcare institutions. The aim was not to decide whether AI is simply “good” or “bad,” nor to compress the issue into one linear theoretical model, but to understand how these experts perceive, adopt, resist, and interpret the methodological and ethical implications of AI in qualitative inquiry. For that reason, the study followed Braun and Clarke’s reflexive thematic analysis and used the Reflexive Thematic Analysis Reporting Guidelines to strengthen transparency and rigor. That methodological choice matters because this article is not searching for a single correct answer; it is examining complex meanings shaped by experience, reflexivity, and positionality (Braun & Clarke, 2021; Braun & Clarke, 2024).

The execution of the study is equally careful. The sample was purposive and included experts from sociology, anthropology, nursing, medicine and surgery, and psychology, all of whom had clear qualitative research credentials, such as completed projects, peer-reviewed publications, formal training, and active qualitative research roles (Campbell et al., 2020). Interviews were conducted in April and May 2025, either face to face or through Microsoft Teams, and researchers also kept field notes capturing nonverbal behavior and interactional cues. All interviews were conducted by an experienced research nurse, transcribed verbatim, and analyzed together with observational memos. The resulting dataset was not justified through data saturation in a mechanical sense, but through sufficient depth and richness to support theme development. The study also received ethics approval from the Regional Ethics Committee of Umbria and handled data in line with GDPR, which means the article establishes its own ethical footing before it begins evaluating the ethics of AI itself.

The findings do not deliver one neat verdict. Instead, they develop four tightly interwoven themes, as if the article were turning the question of “AI in qualitative research” slowly under the light. The first theme examines how researchers perceive the intersection of AI and qualitative inquiry, and the first thing to surface is not enthusiasm but a climate of hesitation, division, and quiet tension. In some research environments, AI is barely discussed openly at all; in others, it is met with visible suspicion, prejudice, or technophobia. Some participants focus on AI’s errors rather than its possibilities, see it as dangerous, treat it as a threat to the essence of qualitative research, or worry that it undermines the researcher’s creativity. What the article captures here is not a simple duel between supporters and opponents, but a genuinely unsettled academic moment: AI has clearly arrived, yet it has not been comfortably welcomed.

An even more revealing layer appears in the subtheme of shame and skill-related barriers. Many participants were not truly avoiding AI altogether; rather, they were reluctant to acknowledge it openly, or only admitted their use gradually as the interview unfolded. Some feared judgment. Some felt embarrassed to disclose AI use. Some realized during the interview that software they were already using, such as NVivo or Lumivero, had incorporated AI functionality. The article’s point here is subtle and slightly mischievous: researchers may already be standing on the threshold of AI while still hesitating to say, plainly, that they have stepped across it. This hesitation is also tied to limited AI literacy. Most participants used general-purpose tools such as ChatGPT and described themselves as largely self-taught, uncertain whether they fully understood AI’s capabilities or limits.

 Human Reflexivity vs. Artificial Intelligence: Can AI Truly Interpret Qualitative Data?

Caution, however, does not equal closure. The third subtheme within Theme 1 shows that a substantial group of participants adopted an open and pragmatic stance toward AI, treating it as a support tool for researchers and as an emerging resource that qualitative inquiry cannot simply pretend does not exist. Some expressed genuine wonder at its capabilities and argued for more openness in discussing it. Yet that openness always came with one non-negotiable condition: human control must remain in the loop. Participants repeatedly stressed critical oversight, data security, integrity, and contextual judgment. In that sense, the more AI-friendly position in the article is not “let AI take over” but “let AI help, while making sure it never gets the final interpretive word.”

The second theme lifts the debate from the level of tools to the level of anthropology and philosophy, and this is one of the article’s richest contributions. Participants drew a strong distinction between machine skills and human capacities. AI may acquire skills through training, but critical thinking, reflexivity, interpretive sensitivity, intuition, creativity, cross-disciplinary linkage, multisensory language, personal experiential knowledge, and contextual immersion were widely regarded as higher-order human capacities that current AI cannot truly replicate. The article emphasizes that the value of qualitative research does not lie only in arranging text neatly; it lies in whether a study can open a new perspective and make the reader respond with something like “wow,” or at least “oh,” because a real interpretive shift has occurred. Without that spark, the research may remain orderly on the surface while losing the depth that makes qualitative work matter.

Even so, the article does not portray the human–machine relationship as a fight to the finish. Many participants described AI as a kind of alter ego, an additional voice, a valid interlocutor, or a tireless research assistant without a soul. For some, the researcher still generates the idea and sets the direction, while AI helps express that idea more clearly or more persuasively. When the dataset becomes very large, researchers may treat AI as another observer and ask, in effect, “What do you see that I do not?” In that sense, AI does not necessarily extinguish reflexivity; it may become an instrument that prompts the researcher to examine their own blind spots more carefully. The picture that emerges is not frivolous techno-optimism but a carefully bounded partnership: AI may sit at the table and speak, yet the researcher still owns the table.

The sharpest warning in this philosophical section appears in what participants called the “illusion of meaning.” The article argues that even when AI is trained and produces polished, coherent outputs, those outputs may only look meaningful without being genuinely interpretive in the qualitative sense. Participants worried that researchers, especially younger ones or those with limited training, might be seduced by fluent language and mistake surface coherence for analytic depth. The concern is not merely that the machine is powerful; it is that the researcher’s own identity may be quietly flattened in the process. When organizational environments prioritize speed, output, and publication, AI can easily serve goals that are measurable and efficient but not necessarily scientific in the deepest sense. What qualitative research most values may be the very thing least likely to be captured by smooth computational output.

The third theme returns firmly to workflow and practice, and here the article becomes wonderfully concrete. On technical tasks, consensus becomes much stronger. Participants broadly accepted AI as useful for transcribing audio, summarizing content, generating visual abstracts or visual models, and especially for handling scientific English. The most widely accepted application was support with English-language writing and translation from Italian into English. The article states this plainly: 13 of the 14 participants reported using AI in some way for English expression. This is not a marginal detail; it is one of the clearest, least contested, and most practically important findings in the study. At the language barrier, AI has already become a very real everyday tool for qualitative researchers.

AI Research Tools for Transcription, Translation, Summarization, and Academic Writing

Once AI moves closer to the core of qualitative research design, data collection, interview analysis, and theme development, agreement quickly loosens. Some participants had never used AI to design a study or analyze interviews; others used it to structure manuscripts according to journal guidelines, generate tailored prompts, or make sense of large-scale discourse data. Yet these deeper applications also drew the strongest skepticism. Participants described AI analysis as a black box, difficult to audit because it is not easy to see why one excerpt is highlighted and another is ignored, or whether the resulting narrative rests on weak premises. The article therefore draws a very precise distinction: AI has earned broad acceptance as technical support, but it has not yet earned equivalent trust in interpretive work.

When the conversation turns to future potential, the tone remains closer to caution than to utopian promise. A few participants imagined that AI might one day facilitate dialogue between methods and disciplines, but many emphasized that the major limitation remains inadequate user training. Advanced, structured learning opportunities are still scarce, so researchers often depend on webinars, peer support, trial and error, and self-directed experimentation. Several participants were also explicit that AI is a poor match for ethnographic research, because ethnography still requires bodily presence, field immersion, and the long cultivation of contextual sensitivity. AI, as one participant effectively suggested, has no body; however quickly it works, it still does not put on shoes and walk into the field. The article does not dismiss the future, but it repeatedly insists that fast technological change does not automatically dissolve forms of qualitative work rooted in embodiment, relationship, and place.

The fourth theme widens the frame again, shifting from method to ethics and sustainability, and this move turns the article into a reflection on what kind of research world AI may be helping to build. Ethical concern is nearly universal across the interviews. Participants worried that AI systems centralize power in the hands of the few actors who design and train them, and that these systems rely on dominant, largely Western epistemological resources that may marginalize minority voices, subcultures, and underrepresented perspectives. At the same time, participants raised hard questions about how personal data are used, who extracts economic value from those data, and whether any benefits are redistributed fairly. The article pushes further by noting that, in the absence of broadly accepted ethical codes, researchers are often left to rely on personal moral reflection alone, and that is nowhere near enough for a challenge of this scale.

Sustainability appears less frequently in the interviews, but the article does not treat it as a side note. Some participants argued that academia pays far too little attention to the environmental cost of AI, including energy demands, server infrastructure, and water-intensive cooling systems. The interviews refer to an estimate that by 2030 the AI economy may consume as much energy as Japan. The point is not merely the number itself, but the question it opens: if AI is imagined as a machine of limitless computational expansion, is the research community also accepting a form of technological anthropocentrism in which computing power outruns ecological responsibility? In the article, sustainability becomes far more than a matter of electricity bills; it becomes a question about cultural responsibility, intergenerational accountability, and the kind of future research is willing to endorse.

AI Ethics in Research: Data Privacy, Bias, Sustainability, and Environmental Impact

In the discussion, the authors place these findings inside a rapidly growing literature on AI in qualitative research and adjacent areas. They note that prior scholarship has already examined AI-led interviews, AI-assisted thematic analysis, AI support for scientific writing, document summarization, and public perceptions of AI in healthcare, while also raising repeated concerns about privacy, reflexivity, trust, and reliability. Compared with those studies, the distinctive contribution of this article is not that it directly tests one particular tool, but that it uses expert voices from Italian academic and healthcare contexts to capture a more culturally and disciplinarily situated form of reflection. AI is not treated here as a standalone technology; it is treated as a social phenomenon that may reorganize research practice, professional anxiety, and scholarly self-understanding.

That is why the article repeatedly argues for a path of balanced integration. Training should not stop at prompt-writing technique; it should expand into critical AI literacy, ethical awareness, methodological reflection, and responsible use. The studies cited in the discussion support that direction from several angles. Some emphasize human–AI “co-intelligence” and collaboration; others call for stronger educational frameworks, clearer disclosure, multidisciplinary teamwork, and more rigorous validation of AI-assisted analysis. The authors therefore do not suggest that every qualitative researcher must become a full AI specialist. Instead, they point toward a more realistic arrangement in which qualitative researchers work alongside AI experts, engineers, or computer scientists, so that these tools can be used more carefully without sacrificing reflexivity or contextual depth.

One of the article’s strengths is how candidly it addresses its own limitations. The sample comes from specific academic and geographic contexts, so the findings do not automatically represent all disciplines, institutions, or regions. AI itself is evolving quickly, which means current perceptions may age quickly too. The study explores subjective perceptions and self-reported experience rather than directly observing what researchers do in practice. In addition, the deliberate decision not to define “Artificial Intelligence” at the start of the interviews, while methodologically useful because it allowed participants’ own conceptualizations to emerge, may also have introduced interpretive inconsistency and reduced comparability across responses. The authors even note one intriguing silence: although AI translation tools are increasingly common in academic contexts, participants did not explicitly foreground them as a separate category, which the article identifies as an issue worth pursuing in future research.

The conclusion therefore does not ask whether AI will replace qualitative researchers. It asks how researchers can refuse to surrender the most essential parts of their work now that AI is already inside the room. The authors conclude that the qualitative research community broadly accepts that technology will continue to evolve and that AI can be helpful for particular tasks. At the same time, participants firmly insist that the human researcher remains central. Multidisciplinary teamwork, careful oversight, critical evaluation, contextual sensitivity, and a sustained “benefit of the doubt” toward the illusion of meaning form the article’s most stable and weighty position. The machine may become faster at transcribing, translating, sorting, and summarizing, and it may even be a talkative assistant at the analytic table; yet in this study, the person who ultimately makes the data speak, brings context into focus, and gives interpretation its legitimacy is still the researcher.

References:

Braun, V., & Clarke, V. (2021). Thematic analysis: a practical guide. SAGE.

Braun, V., & Clarke, V. (2024). Supporting best practice in reflexive thematic analysis reporting in Palliative Medicine: A review of published research and introduction to the Reflexive Thematic Analysis Reporting Guidelines (RTARG). Palliative Medicine, 38(6), 608–616. https://doi.org/10.1177/02692163241234800

Campbell, S., Greenwood, M., Prior, S., Shearer, T., Walkem, K., Young, S., Bywaters, D., & Walker, K. (2020). Purposive sampling: complex or simple? Research case examples. Journal of Research in Nursing, 25(8), 652–661. https://doi.org/10.1177/1744987120927206

Dellafiore, F., Saba, A., Collaro, C., & Artioli, G. (2026). Artificial Intelligence in Qualitative Research: Insights From Experts via Reflexive Thematic Analysis. Qualitative Health Research, 36(2–3), 145–165. https://doi.org/10.1177/10497323251389800