Qualitative Research in the Age of AI: Considerations, Critiques, and Craft

For a brief period in November 2025, the number one country song in the United States (by digital sales) was “Walk My Walk” by Breaking Rust. What’s interesting is that everything about Breaking Rust and their music—the composition, production, and even the cowboy imagery in the accompanying music video—was AI-generated. By some accounts, there is no Breaking Rust.

For some of the millions of people who enjoy listening to “Walk My Walk,” this caveat may not matter. A good song is a good song. Indeed, in the YouTube comments that accompany the song, one person writes, “Don’t care who made it. It is a kicking song.” Another writes, “It’s about where a track takes you and if it connects with you. Music is music in any format…it’s how the song makes you feel.”

For others, however, the reaction was more tepid. One comment read, “So soulless the music with machines. [S]o stupid.” Another read, “That’s not music… It’s AI diarrhea. [The] problem is some people like AI diarrhea.”

Still other commenters jumped straight to the end game. “We are absolutely COOKED,” read one reaction. Another commented, “Oh man…sad we don’t even need people for any jobs anymore.”

Research articles in journals such as Organization Science aren’t exactly like country songs. No one is dancing to our process models or singing along to our theoretical insights at the local bar. Yet the debates about the use of AI in both music and research have parallels. Just as questions, excitement, and concerns about AI use have rocked the music industry, they’re also rocking the world of academic research. Look no further than the analysis of submissions to this very journal, published earlier this year, which shows submission volume is up dramatically since the release of ChatGPT in November 2022. (That analysis cited a 42% increase though more recent figures show a 54% increase in just the past year). Lower quality, AI-generated writing accounts for nearly the entire increase, and qualitative submissions experienced an even larger increase than quantitative ones (Gartenberg et al. 2026). As qualitative scholars, where does this leave us?

To some extent, our field’s varied responses mirror the YouTube comments on “Walk My Walk.” Some scholars contend that what matters is the end product (in our case, the published paper) and that the process used to create it is secondary, if not wholly unimportant. Others focus on the quality of current AI-assisted or AI-generated scholarship, arguing that it simply does not match the best of what human scholars can do. By extension, these scholars expect AI to make qualitative research more widely available and practiced, but maintain the belief that trained human scholars retain unimpeachable expertise in doing this work. Still others look ahead, asking where continuous technological improvements may lead and whether human scholars will still have a role in the future, and if so, what? And, of course, there are many other views, too.

A sampling of these different reactions can be found in comparing two recent reflections in Organization Studies. In one reflection, Henri Schildt (2026, p. 1) acknowledges the pitfalls of AI before asking, “Can management scholars decide to sit this out?” (The “this” in question is the use of AI in our knowledge work). Schildt (2026) ultimately makes a case for using AI in qualitative research, pushing for “responsible” use by which researchers can examine the documents on which AI formed its interpretations as well as those it chose to leave out. Schildt’s (2026) arguments rest on the premise that we have access to the tools, and other knowledge workers such as lawyers and physicians are already making use of them for “lay qualitative research.” Therefore, as scholars, we should experiment with them and figure out how best to incorporate them in our research processes and workflows.

In another reflection, Cornelissen (2026) argues that the use of AI tools will commodify theory, creating “many more simulated theoretical arguments, frameworks, and explanations that are false positives, industrial-scale data mining, and HARKing (hypothesizing after the results are known)” (p. 2). In other words, we’re already awash in theories, constructs, and mechanisms. Do we really need AI generating more of them, more quickly? Citing Cartwright (1999, p. 247), Cornelissen (2026) likens this possibility to a “vending machine” in which “you feed [the AI tool] input in certain prescribed forms for the desired output; it gurgitates for a while; then it drops out the sought-for representation, plonk, on the tray, fully formed, as Athena from the brain of Zeus” (Cornelissen 2026). Cornelissen (2026) ultimately calls for us to avoid AI-augmented theory building, emphasizing the critical role of humans in probing rich qualitative data and making sense of emergent inferences. Here, Cornelissen’s (2026) arguments are based on the premise that there does not exist a functional equivalence between algorithm-based computational tools and embodied human beings. Thus, why should we offload theory building to machines that lack a contextualized understanding of the world? Such concerns have led some scholars to reject the use of AI tools for reflexive qualitative research altogether (e.g., Jowsey et al. 2025).

The result is a building debate between two views, one more fully embracing AI in qualitative research and the other largely rejecting it, with both resting on distinct warnings and visions of the field and the future. Whereas Schildt (2026) warns of obsolescence with key stakeholders if we fail to adopt AI, albeit responsibly and with guardrails, Cornelissen (2026) warns of a looming credibility crisis if we do adopt it, at least for some tasks such as theory building. In the midst of these positions (and others), scholars are seemingly forced to select between obsolescence and a loss of credibility, hardly an enviable position.

Part of the challenge, of course, is that there are many approaches to qualitative research, many applications of AI for qualitative research, and many arguments to be made for and against these applications. Yet, amid these perspectives and arguments, what is sometimes lost is space for epistemic pluralism and for deep, critical thinking about the broader issues surfaced by AI. It is in this space in which we, as qualitative editors at Organization Science, seek to offer our perspective. Our view is not the journal’s policy, and there may be differences of opinion among the journal’s qualitative editors, not to mention the journal’s editorial team as a whole. Such differences are healthy. We offer this perspective with the hope of advancing important conversations taking place about this technology and raising considerations for qualitative researchers, in particular, to contemplate.

We proceed by reviewing some of the claimed advantages for AI use in qualitative research. We then consider some of the concerns surfaced by such use. Finally, we raise several questions for the qualitative research community, and the research community more broadly, to ponder. Central to our comments are the tasks that AI might perform, questions of epistemology and of “output” versus “process,” and larger systemic considerations. Ultimately, we reflect on what may be the biggest questions that AI forces us to confront: what’s the purpose of scholars and scholarship anyway?

Some Claimed Advantages of Using AI for Qualitative Research

The term “qualitative research” covers a wide range of approaches and data sources, including ethnographies, multiple case studies, in-depth interviews, and discourse and archival analyses. When considered together, what these approaches have in common is a reliance on and a deep relationship with language and writing as core to the research process. These very characteristics also seem to make qualitative research an ideal match for large language models (LLMs).

Many researchers and AI providers tout AI as a labor-saving tool that increases productivity by enabling researchers to go faster. Initial applications included tasks such as scraping archival documents from electronic repositories and transcribing interviews (e.g., Otter.ai). But, clearly, some researchers and tool providers have much grander aspirations. For example, the marketing materials for Atlas.ti claim that AI integration “removes the headache from qualitative data analysis and lets the software do the heavy lifting” and emphasize Atlas.ti’s ability to “generate deep insights automatically, leveraging the latest AI and machine learning algorithms for faster results.” Not to be outdone, NVivo claims, with its latest release, “The standout feature of NVivo 15 is its groundbreaking Lumivero AI Assistant. This powerful tool is designed to enhance data analysis efficiency and accuracy.”

Of course, AI can do more than analyze qualitative data. Tools such as Quallie can generate data through “synthetic respondents,” though Quallie also offers the following disclaimer: “Use it only for testing your materials and training purposes.” Although there may be reasons to pilot an interview protocol with Quallie before talking to human respondents, it may also become increasingly difficult to distinguish between synthetic and “real” data (and institutional review board protocols can add to the challenge of detection because they limit what might be shared). In a world that places a premium on speed and in which a single “traditional” interview might take a full day to arrange, prepare, and conduct, it may be tempting to use AI tools that can generate dozens of synthetic interviews in a matter of minutes as a primary data source.

Moving beyond collecting, generating, and analyzing data, AI can also write up findings and position these findings within an AI-generated literature review. Indeed, even generic LLMs such as Claude Opus 5 can now generate convincing manuscripts, complete with citations that are increasingly “real” and not hallucinated. In turn, scholars can readily generate entire papers in less than a day—an inconceivable feat just a few years ago. (As an extreme example, Cornelissen (2026) cites the example of Novy-Marx and Velikov (2026), who generated 380 economics papers in a single day, though it is worth noting that they were not qualitative). With continued improvement in AI tools, the rate of production will undoubtedly increase, even as the requisite skill of the user will decrease. The promise is, thus, a world in which AI enables scholars, including those with less training and experience (or, really, anyone), to produce more qualitative papers faster.

Though why would we want to generate more qualitative papers and do so more quickly in the first place? As highlighted by Organization Science’s analysis of submissions in the age of AI, this desire for enhanced productivity is tied to incentives and rewards at all levels, from PhD students seeking their first academic job, to tenure-track faculty trying to attain the magic number of publications required by their school to get tenure, to tenured faculty members chasing course releases, summer support, and financial rewards for publishing in the “right” outlets (Gartenberg et al. 2026). At every level, a key signal of one’s value and success as a scholar seems to be the number of publications produced. Because qualitative research, especially, is known for taking a long (long!) time, a tool that promises to speed up the process holds obvious appeal.

Taking this logic a step further, AI tools could, in theory, even push more scholars toward qualitative research approaches. When faculty (and PhD students) weigh research approaches, the rational choice might now be to opt out of qualitative work simply because it takes too long and, thus, disadvantages them in a job (or PhD program) in which the tenure clock (or dissertation timeline) offers a finite amount of time regardless of methodological approach. As such, AI-supported qualitative research could enable more scholars to pursue qualitative research, perhaps a positive outcome.

Moreover, even though the key sales pitches for using AI in qualitative research seem to be increased productivity and efficiency, other justifications also emphasize improved quality. For instance, some qualitative scholars use AI as a “sparring partner” in tasks such as contemplating construct labels and generating examples to assess boundary conditions. Especially for scholars who lack qualitative colleagues or robust academic networks, this use could be advantageous if their choice is between AI and nothing. Others may find value in using AI to help with a literature review by surfacing overlooked articles or perspectives. In fact, in a world of rapidly proliferating journal articles, AI may be the only way to review all the relevant prior research. (At the same time, AI tools also tend to miss articles behind publisher paywalls, resulting in incomplete and/or vacuous literature reviews; Poe and Dryer 2026.)

If marketing materials are to be believed, AI may also enable more accurate coding (recall NVivo’s claims that it enhances efficiency and accuracy). Yet, here, optimistic marketing claims begin to run headlong into the epistemology of qualitative research. Specifically, accuracy is often tied to a certain perception of objectivity and to measurements that reflect an objective reality. Certain applications of qualitative research, such as market research conducted by firms, focus here, too. For instance, a researcher may wish to capture objective characteristics (e.g., “How many times did each respondent mention ‘alpaca’ in their interview?”) or perhaps pseudo-objective perceptions (e.g., “Was the valence of this passage positive or negative?”). However, many qualitative researchers, and especially those attempting to publish theory-oriented articles in journals such as Organization Science, adopt interpretive and constructivist approaches in their work, their insights relying on subjective interpretations rather than objective(-ist) counting. At the very least, these approaches suggest that claims of AI coding “accuracy” need to be interrogated against epistemic ideas of “objectivity” (e.g., Daston and Galison 2021) and reflections on the purpose of qualitative research. As such, these considerations bring us to some broader concerns associated with using AI in qualitative research.

Some Concerns With Using AI for Qualitative Research

Many concerns with the use of AI for qualitative research stem from the onto-epistemological stance of many qualitative traditions. Qualitative researchers often subscribe to a view in which reality is situated in time and space within a sociomaterial context; particular realities are socially constructed through norms, rites, and rituals; and people are active sense makers (meaning that qualitative researchers must understand their informants’ perspectives and contexts, not superficially, but deeply and meaningfully). In turn, the point of much qualitative work is not to uncover a universal truth or singular and objective reality so much as to convey a compelling analysis and interpretation that may be insightful and used as a basis for further inquiry. Such interpretations rely on the researcher’s human experiences and contextualized understandings of the world. Though different qualitative researchers may hold differing views on the extent to which the world is socially constructed, the typical appeal of qualitative approaches lies in the desire to capture phenomena that cannot be (easily?) measured and counted, and that require attention to nuance, social context, and interpretation.

Viewed from this perspective, claims of AI accuracy and, thus, the presumed elimination of variance stemming from inaccuracy are odd. That is, human-induced variation and noise may not be “bugs” to be corrected through AI’s enhanced accuracy but rather features that lead to more interesting work, particularly when richly captured in a qualitative study’s methods section (Glaser and Strauss 2017). By contrast, widespread AI use in qualitative research may erode useful variation that leads to important insights. In a recent article in Science, Evans and Duede (2025) warn of this very possibility for research in general, writing, “AI’s efficiency could create a monoculture of research practices within and across fields.”

This same onto-epistemological stance means that qualitative research places more emphasis on process than on output—an emphasis that sits in direct contrast to AI, which focuses largely on outputs. Indeed, even the designers of leading AI tools do not fully understand the details of the processes that generate a given output. Yet process is paramount, not only in qualitative research, but in research and creative activity more generally. For example, a recent editorial in Nature details how the process of writing is linked to thinking, with the former supporting the latter (Nature 2025, see also Fayard and Metiu 2012). Conversely, researchers have found that LLM use is associated with a lack of critical thinking (Kosmyna et al. 2025), as evidenced in humans’ adoption of AI outputs with little to no scrutiny (Shaw and Nave 2026). Echoing these same points, a recent essay in the Chronicle of Higher Education by a creative writing professor claimed, “Speed is the enemy of thinking. …Creative writing, at its best, is inefficient. It requires false starts, dead ends, bad metaphors you have to live with long enough to recognize them as bad” (Pittard 2026).

Perhaps the same might be said of AI and qualitative research: qualitative research, at its best, is inefficient. It requires false starts, dead ends, and empty constructs you have to live with long enough to recognize them as bad. Indeed, many qualitative researchers have the (frequent!) experience of struggling to make sense of something or seemingly hitting a wall, and then going for a walk, reading a novel, or listening to an album only to find an unexpected connection or pathway forward. A key question then is whether eliminating this friction and tension in the interest of speed and efficiency may, in fact, make for lesser insights. As music producer Brian Eno said in a 2025 interview with Ezra Klein, “Friction is very important. Friction gives you a little time to see what’s happening. It makes something ease into your life more slowly so you can start to correct it as it’s easing in.” Expressing a similar sentiment 60 years ago, when confronted with the use of computers to speed up the music composition process, composer John Cage commented, “What we need is a computer that isn’t labor-saving, but which increases the work for us to do.”

Writing not only is a metaphor for the process and craft of qualitative methods but also is a central activity in any qualitative research process. Ethnographies, interview-based research, discourse analysis, and archival studies all involve writing field notes, memos, and/or diaries (e.g., Van Maanen 1995). In each case, qualitative researchers engage in interpretation and textual construction through the process of writing.

Here, some might think that using AI to write field notes, memos, and diaries, which researchers would otherwise craft manually as they observe and reflect, is exactly where these tools can be of great help. Using AI to write (or heavily edit) a manuscript might seem appealing, too, especially for non-English speakers. That way, researchers can focus on what seem to be more human or value-added tasks. Yet there are consequences to using AI to produce text in place of human writing. First, the use of AI for tasks as seemingly benign as editing may end up changing intended meanings. In a recent study, researchers asked three production LLMs (GPT5, Gemini 2.5, and Claude Haiku) to revise a set of human-generated essays. Even when they used expert prompts and told the LLM to make only grammatical edits, they found that LLMs “not only alter the voice and tone of human writing, but also consistently alter the intended meaning” (Abdulhai et al. 2026).

Second, because of the deep connection between writing and thinking, letting an AI assistant do the writing may also mean sacrificing the thinking (recall the creative writing professor cited above). In a recent New York Times Magazine article, a doctor reflected on her evolving view on the use of AI scribes to write clinical notes. She initially hoped such scribes would be more efficient and enable her to focus on her patients, just as some qualitative researchers might hope that AI will free them to focus on higher-value tasks. But the doctor's perspective shifted after continued use and reflection:

For a while, after I started using the scribe, my role in creating the medical note seemed simple: read the draft, correct what was wrong, sign it. So fast, so easy. But then the process began to discomfit me, as I slowly realized that editing a note I did not create does not demand the same of me as writing a note. I was no longer thinking through the interaction with a patient and the meaning of the information I had gathered from the visit. I was checking a version that had already been made for me. This distinction seemed small at first. The notes had been generated from my encounters, from the things that were said in the exam room. And I was still revising and proofreading the A.I.’s output afterward. Over time, though, I have come to see how much of my own thinking had been bound up in the writing process itself (Ouyang 2026).

Use of the AI scribe not only altered the doctor’s writing of clinical notes but also changed her interactions with patients and her deliberations around diagnoses because she did not engage in the same thought process. She realized, much as John Cage realized 60 years ago, that she did not necessarily want a technology that was labor saving if it meant losing the ways of thinking and interacting that make her a doctor.

Even as the relationship between writing and thinking may be among the most intractable issues surrounding the use of AI for qualitative research, other concerns, too, merit consideration. For instance, the use of black-boxed AI tools can run counter to transparency about the research process. Transparency around process is important to readers and reviewers alike as they seek to understand a study’s findings. Indeed, a common reviewer comment on initial qualitative submissions is, to paraphrase, “You need to explain your analytic process in more detail so that we understand how you moved from your raw data to the insights that you present in your findings.” The answer of, “I’m not sure because I followed Atlas.ti’s marketing materials and let AI ‘do the heavy lifting’” seems unsatisfactory to us. Indeed, it is notable that, in the social sciences more generally, we have had vigorous calls for transparency (e.g., Schwab and Starbuck 2017), encouraging the publication of not only results but also data and analyses that open the black box of the processes that led to those results. Although there are many reasons why full transparency by qualitative scholars can be inappropriate, such as adhering to privacy protections for interviewees (cf. Pratt et al. 2020), it remains striking that in the midst of broad calls for transparency, we are also seeing increased use of analytical tools that are inherently black-boxed.

The potential for AI to obscure the processes behind published insights can be detrimental to the training of future scholars, too. Specifically, if these scholars do not learn—in detail—the processes needed to generate credible insights in qualitative research and they instead outsource these processes to a black-boxed technology, they may well be disadvantaged as they attempt to progress and grow. An apt analogy might be found in Callen Anthony’s (2021) study of investment bankers who relied on black-boxed software tools. Anthony (2021) finds that some junior bankers blindly trusted the output of the tools, whereas others worked to understand how the tools generated particular outputs. Although both sets of junior bankers succeeded in the short term, those who worked to understand the “how” had better long-term career prospects; after all, they learned how to think and how to apply this analytic ability in ever-new settings, including those untamed by software tools. In contrast, those who blindly applied the software tools faced upward mobility challenges and many exited the profession altogether. By extension, the use of AI tools absent deep understanding may enable a PhD student to complete a qualitative dissertation faster, but this same use may inhibit the development of skills that set one up for long-term success as a scholar. Indeed, especially in high-skill and creative fields, success often depends on leveraging expertise above and beyond what’s available to anyone with the same set of automating tools (Nelson et al. 2023).

The use of AI for qualitative research raises other challenges, too. For instance, AI coding tools can miss contextual nuances or other clues that are important in developing research insights. As an experiment in 2025, a PhD student at the University of Oregon used Perplexity.ai Pro to automatically code the sentiment of various media posts on cryptocurrency. He knew he had a problem when the software coded the following passage as positive sentiment: “Trump, the most trustworthy businessman, is teaming up with crypto, the most trustworthy business.” Even today, sarcasm detection remains difficult for some AI tools. In more complex settings, such as ethnographies, it is essential to understand the setting to appropriately code what happened, when it happened, and why and how it happened. Even when scholars attempt to use AI to code their field notes, these codes often are not informed by the scholar’s tacit understandings. It might be as one of our colleagues told us after experimenting with an AI tool for coding, “Can we agree that these tools just aren’t very good for the type of work qualitative coding demands?” Echoing our reflections on the writing process, it is worth noting that the answer is not simply waiting for the technology to get better, which it surely will; it also demands consideration of the very human and personal interpretative process that takes place during coding.

Yet another concern is that the use of AI may shift the composition of qualitative research approaches. For example, current AI models are quite good at assembling and reviewing text and even videos. Thus, AI may may drive qualitative researchers to conduct studies that draw on such data. Yet much material of interest does not exist in tidy digitized formats. Ethnographies, for example, are a staple of qualitative research, providing insights that archival documents and other data sources simply cannot (e.g., Van Maanen 2011). It is difficult to imagine an AI tool conducting an ethnography. A key question is, thus, whether the use of AI may push qualitative research in directions that are better fits for the technology and, conversely, lead to a decline in qualitative research approaches that are uniquely insightful but worse fits for the technology.

Even the use of AI as a sparring partner may carry risks, such as the loss of community and dialogue, which are central to the academic enterprise. This loss is an issue not only from a human and social perspective but also from a critical thinking perspective. For instance, a recent Science study showed that LLMs are overly agreeable when it comes to interpersonal matters (Cheng et al. 2026). Could it be the same with analytical interpretations? As in other parts of the research process, the use of AI as an intellectual partner, thus, raises important questions regarding how much we trust AI interpretations (versus our own) and how much we outsource to AI (e.g., what is core to the research process, and what is not?).

Some Outstanding Questions

Both the possibilities and concerns outlined above raise numerous questions. First, AI clearly enables less skilled or less experienced researchers to generate scholarly output in the same way that it enables someone with little artistic skill to create an image or someone with little musical skill to compose a song. In this way, AI raises the floor of performance in qualitative research. However, we know less about what it does to the ceiling. One possibility is suggested in the Nelson et al. (2023) study of technology evolution in the music synthesizer industry. They find that digital synthesizers offered a seeming advantage over their analog predecessors because they made it easy to access new sounds. Yet they did so by black-boxing the sound-generating process, thereby preventing skilled users from fully leveraging their expertise. In this way, digital synthesizers raised the floor for amateurs, but they also lowered the ceiling for experts. In other environments, however, we might imagine new technologies raising the floor and leaving the ceiling unchanged or raising both the floor and the ceiling.

One key question for qualitative researchers as they encounter AI is which of these scenarios will hold. Moreover, even if AI does lower the ceiling, is that a trade-off worth making for the gains that come from raising the floor? That is, should we be more concerned about how AI shapes the most insightful and influential qualitative research (i.e., the type of work we aim to publish in Organization Science) or how it may enable a much wider array of scholars to engage in qualitative research, period? The question is further complicated by the fact that there are, by definition, more amateurs in any field than there are experts, thereby resulting in greater market interest (e.g., on the part of AI tool providers) in raising the floor than in tending to the ceiling.

Second, how might the production, review, and consumption of scholarly research respond to the increasing use of AI? One strong possibility is that AI use begets AI use. As noted, AI enables us to produce more papers faster. Even with current-generation technologies, it’s not difficult to imagine a fivefold increase in a scholar’s productivity. Indeed, we have recently had authors submit as many as five unique papers to Organization Science in a six-month period, openly using AI in almost every aspect of the research process. Where does this lead us? One answer is that we need five times as many editors and five times as many reviewers (or we need the existing pool of editors and reviewers to do five times as much work). The natural response would be for these individuals to use AI to write reviews and to synthesize reviews and to write decision letters. Indeed, the Gartenberg et al. (2026) analysis of Organization Science submissions shows that this is already beginning to happen with reviews. (On the editorial side, the response, which is unsustainable, has been to add editors and to dramatically increase editors’ workloads.) In turn, AI could be used by authors to respond to decision letters and by reviewers to review resubmissions and by editors to write new decision letters and so on and so on. Ultimately, we could have five times as many articles published and then 10 times, 20 times, and 50 times as the technology improves. But where does this leave us? Will we, as the humans in the loop, develop a meaningful understanding of any of them? And will they, as the papers being produced, have any meaningful impact beyond the mere fact that they exist?

Most scholars already feel overwhelmed by the number of publications and comment that they have too little time to read them all. As Bechky and Davis (2025, p. 3) lament in a recent piece in Administrative Science Quarterly, “[D]eep intellectual engagement can increasingly seem like a lost art in the current career and publication environment. In a system driven by quantitative metrics, the point is not to be read but to be published and cited. Like unworn articles of polyester clothing piling up by the acre in Chile’s Atacama Desert, unread research articles pile up by the thousands in online open access journals.”

With a fivefold increase in qualitative publications, largely driven by AI, scholars may have no choice but to rely on AI to read and interpret this work, too. And so the cycle goes (and the discourse of inevitability churns). Followed to its logical conclusion, it is not difficult to imagine largely removing humans from research and publishing altogether. Then, some key questions include, “Is that good for scholarship?” and “Is that good for scholars?” Of course, these questions may not have the same answer, nor are they unique to qualitative research.

A third and related question concerns the end game. Academia is based on reputations and, relatedly, rankings (whether implied or discrete). If AI eventually makes it possible for anyone regardless of skill or training to produce papers, again raising the floor, then paper production (i.e., a count of outputs) may lose meaning as a source of differentiation between scholars or programs. Put slightly differently, articles published in Organization Science and similar journals may be valued, in part, because they are relatively rare compared with the entire corpus of management research. But, if it becomes easy to generate such articles, then the rarity and, thus, the perceived value may diminish. That may be a good thing and may nudge us to consider more than just counts when assessing scholars and scholarship. However, such a shift would also require an overhaul to our current incentive structures that claim to desire quality but that measure volume, with additional weight given to more outputs in the “right” outlets and relatively little consideration given to the process used to generate those outputs.

Finally, a fourth question, and one that lies at the heart of our reflections, is whether and how we care about the process (including the people) used to generate research or only the output (i.e., published papers). Much of the rhetoric around AI use in research focuses on research outputs. For instance, Administrative Science Quarterly’s AI guidelines note, “We worry about a future in which AI assistance makes it ever easier for authors to churn out work that does not meet the field’s standards” (Administrative Science Quarterly 2026). To be sure, the quality of AI output is a key concern for reasons we touch on above. Indeed, some observers liken AI-produced scholarship to intellectual fast food: it tastes good and it satisfies, but it’s not healthy, and it will kill you if it’s all you consume. Again, the Gartenberg et al. (2026) analysis of Organization Science submissions is informative here, showing that quantity has increased but quality has not (yet).

But what about a future in which AI churns out work that does meet the field’s standards? Or, by analogy, produces fast food that is just as healthy? Especially in light of the rapid pace of technological change, if we attempt to differentiate human- and AI-generated content on the basis of perceived quality of the final product, the gap is shrinking and may well disappear altogether. Then what? Much AI-generated work already is difficult to distinguish from human-generated work as with some AI-generated images, some AI-generated text, and some AI-generated music. (Indeed, some of the many comments about Breaking Rust’s “Walk My Walk” indicate surprise, as with “Wait. This is AI?!”)

Food, of course, is about both the product and the process. Imagine, for example, that we had a microwave that could quickly and easily produce meals nearly identical in final form to something home-cooked. Would dinner party guests feel cheated by someone serving a microwaved dinner? What about the experience of the person producing the dinner? Turning again to a musical analogy, will audiences flock to see Breaking Rust in concert? And if they do, will they enjoy cheering a laptop or a hologram on stage? Or, as yet another example, if one is standing in Madrid’s Museo Reina Sofia and looking at Picasso’s Guernica, would the piece have the same impact if AI had produced it? Or does it have impact because Picasso spent more than a month painting it, reflecting on his lived experience with Franco, the joint Nazi-Italian bombing of the Basque Country, and the Spanish Civil War? As qualitative researchers, we must likewise ask if we care about process and creators and the unique lenses each of us bring to our contexts or only the final product.

In the music industry, among others, some observers are clinging to this emphasis on process. For example, a blogger for the record label MAD Records wrote, “The AI wave is coming, and there is no stopping it… Artificial intelligence will likely dominate the mass music markets… AI will know what we want more than we do, and in that regard, it will certainly replace large parts of the music industry. But I believe the AI revolution will also reveal a hidden truth about music (and art in general) that many of us have forgotten in the last two decades…that art isn’t about the final product at all, but rather the struggle to create it” (Gilbride 2025).

There are two parts of this commentary that we might reflect on as qualitative scholars. The first is, again, a perceived inevitability to AI: “The AI wave is coming, and there is no stopping it.” Of course, this is a common refrain when it comes to AI and higher education, too (see, e.g., Kraynak 2026). As such, it can also become a self-fulfilling prophecy: if we think it is inevitable, then it very well may be. However, scholars of technology have long eschewed such determinism (e.g., Bijker et al. 1987, Nelson 2024). To be sure, engineers make design choices that foreground certain technological features, billion-dollar companies have considerable power in pushing their tools, and governments deploy technologies in ways that advance the objectives of the state. Yet new technologies such as AI require integration with institutions—culture, norms, organizational models, occupational incentive systems, and so on—to have certain effects. Whereas scholars may feel powerless to change AI models, they are certainly not powerless to change the associated institutions. Moreover, there are smaller and more invisible ways we can take agency, both individually and collectively: in deciding not to use certain tools just because everyone does; in continuing to read, write, and do the work, even if it’s not always frictionless (especially if it is not frictionless?!); and in nurturing a community of scholars with whom we engage intellectually and support emotionally.

The second part of the MAD Records commentary asks, “What’s it all about?” At least in music and at least in the opinion of one record label, it “isn’t about the final product at all, but rather the struggle to create it.” Turning to qualitative research, we too might ask, “What’s it all about?” In answering this, we might consider two valences of the question. First, what are scholars all about? What is the point of qualitative research for each of us as individual scholars? Why do we do it, and what do we hope to get out of it? Is it a means to an end of a longer CV? If so, then if we can speed up the means by using AI, then we should use it as much as possible at every juncture imaginable. It would only be rational. But for many of us, such as the artists at MAD Records, the process itself is part of what fuels our engagement and identity. As David Bryne of the Talking Heads noted, “We don’t make the music—it makes us.” Herein lies the friction and tension, the false starts and dead ends, and ultimately the rewards of engaging in the inherently human processes of interpreting, understanding, and creating. In turn, we need to ask if automating certain tasks—even if they’re slow, and frustrating, and a struggle—may be a Faustian bargain. Put slightly differently, if the processes of exploring a phenomenon, asking questions, developing interpretations, rethinking our interpretations, and writing are neither fast nor easy, then perhaps that’s why and how they provide meaning.

Second, what’s the point of scholarship—beyond ourselves—for our scholarly community, for the management community, and for society writ large? Why do our colleagues, universities, governments, and neighbors support our scholarship, and how do these groups hope for us to approach our research? Do they care only about output or also about process? Of course, there are certainly many answers. Yet before we (cynically?) settle too far on the view that process may matter to creators, but output is most important to consumers, we might consider the example of craft production. From many studies of many types of craft, we know that there are domains in which producers and consumers alike care deeply about process and not just outcome. For instance, even if one can buy a mass-produced sweater that is more durable and uniform than a hand-knit one (and cheaper, too), the very fact that someone hand-knit a sweater can make it more valuable. At the same time, the vast, vast majority of sweaters sold and worn are mass produced. These differences reflect different goals on the part of different consumers: some place an emphasis on price or utility (“I’d like a reasonably priced sweater that looks nice and will keep me warm”), whereas others place an emphasis on uniqueness and on connection with a creator and their process (“I watched a Peruvian craftsperson knit this sweater with a unique pattern, using wool from an alpaca that her family owns.”) A key question, therefore, is why and how our communities and supporters value our research and whether the process used to generate it (e.g., AI) shapes this value. Of course, collectively we are our communities, and each of us is both a producer and a consumer of research. Thus, far from AI carrying a certain inevitability, our individual choices, guided by our own values, can be expected to determine how this technology (or, really, any technology) shapes our research and its perceived worth.

So what does all this mean? Returning to the Breaking Rust example, whether you enjoy “Walk My Walk,” are discomfited by it, or wonder what it means for the future of music, two things seem to be true. First, the process of creating a song—to be clear, the artist’s process of creating a song—encompassing experiments, unexpected sources of inspiration, and lived experiences cannot be replicated by a technology. Second, the artist’s evolving expression of a song, comprising live performances, remixed versions, and the memories and emotions listeners associate with it, constitutes a uniquely human-to-human phenomenon that cannot be manufactured. We believe these truths generalize to the craft of qualitative research. In our world of knowledge production, as technologies continue to evolve and researchers continue to experiment with them, what may come to matter most is not the outputs at all but rather the processes that underlie them and the ways in which their socialization may advance the field, inspire future generations of inquisitive thinkers, and perhaps even impact society more broadly.

Ultimately, the questions about AI and qualitative research are not only or even primarily questions of technology implementation or even of automation versus augmentation of different tasks. Instead, they’re questions of institutions and incentives that drive and support our work, of the emphasis we place on process versus outcome, and, fundamentally, of occupational identity for qualitative scholars: who are we, what do we do, and why?

Perhaps fittingly, these questions would make for a great qualitative study.