Evidano home
Published in Frontiers in Research Metrics and AnalyticsResponsible AIKathmandu University School of Education, Lalitpur, Nepal

Rethinking Rigour for AI-Assisted Qualitative Research: From Crisis to Collaborative Criteria

Evidano7 min read
Niroj Dahal, Kathmandu University School of Education

Featured researcher: this work is featured independently and does not evaluate Evidano.

A qualitative researcher in Kathmandu, Dhaka or Terengganu now has tools that can draft interview guides, suggest codes and summarise transcripts in seconds. The questions this raises are not only practical. Whose voice is speaking when an AI tool helps write up participants’ stories? Can readers trust claims shaped by models trained on incomplete and biased data? And who participates in change when an AI system becomes part of a participatory or community-based project?

Those questions frame “Navigating the meta-crisis of generativity: adapting qualitative research quality criteria in the era of generative AI,” a perspective article published in Frontiers in Research Metrics and Analytics in 2025 by Niroj Dahal of Kathmandu University School of Education, with Md. Kamrul Hasan, Amine Ounissi, Md. Nurul Haque and Hiralal Kapar. Drawing on the authors’ own experiences and a review of the literature, the article argues that generative AI does not so much create new crises for qualitative research as intensify three old ones, and it proposes practical ways to adapt quality criteria so that researchers can use these tools without losing the humanistic foundations of the field.

We spoke with Dr. Niroj Dahal of the Department of STEAM Education at Kathmandu University School of Education, to learn more about the framework and how researchers can put it into practice.

About the article

Research question

The article asks three linked questions: how researchers can keep their own reflexive voice visible when AI tools may replace or distort their understanding of participants’ stories; whether research that uses opaque, potentially biased AI systems can still be trusted, and what new checks are needed; and how to respect communities’ decision-making and agency when AI is used as a collaborator in research with them.

Key finding

The authors conceptualise a “meta-crisis of generativity”: a convergence of Denzin and Lincoln’s crises of representation (“who speaks?”), legitimation (“can we trust this knowledge?”) and praxis (“who participates in change?”). They argue that this meta-crisis is “not a threat but an invitation” to reanimate the critical ethos of qualitative research, and they propose strategies and adapted quality criteria, such as algorithmic accountability and human-AI dialogue journaling, that let researchers harness generative AI while keeping human accountability at the centre.

Why this research matters

Debates about AI in qualitative research often swing between excitement about efficiency and outright rejection. Much of the discussion also comes from North American and European institutions. This article reframes the debate through the long-standing language of qualitative rigour, asking what credibility, transferability, dependability, confirmability and reflexivity should mean when a non-human tool participates in analysis and writing.

It also brings a Global South perspective, written by scholars in Nepal, Bangladesh and Malaysia, and calls for culturally attuned AI tools and for decolonised training data to prevent epistemic harm. For researchers who want to use AI responsibly but lack practical guidance, the article translates abstract concerns into concrete practices that can be documented and reviewed. Niroj Dahal explained the motivation behind the article: “Since ChatGPT arrived in late 2022, my co-authors and I have watched generative AI enter almost every stage of qualitative work, from drafting field notes to generating themes. The questions it raised felt familiar: who is speaking, can we trust this knowledge, and who takes part in change? Those are the crises of representation, legitimation, and praxis that Denzin and Lincoln named decades ago. We saw them converging into what we call a meta-crisis of generativity, and we wanted to give researchers a way to work through it.”

Key findings from the article

Three contributions stand out for researchers deciding how to use generative AI in their own studies.

  • Representation: keep the researcher’s voice visible. To prevent AI-generated text from silencing the researcher’s reflexive voice, the authors propose voice annotation and reflexive bracketing, and co-authorship frameworks that credit tools such as GPT-4 as non-human collaborators rather than hiding their role.
  • Legitimation: make AI use auditable. The article sets out five strategies for trustworthiness: algorithmic audits with data scientists; triangulation that cross-checks AI outputs against independent human coding, multiple tools, member checking and thick description; audit trails of every AI interaction and prompt iteration; active bias mitigation; and validity checklists adapted to include AI transparency metrics.
  • A GenAI-adjusted rigour framework. The authors pair each traditional criterion with an adjusted one: credibility with algorithmic accountability, transferability with contextual embedding logs, dependability with system transparency documentation, confirmability with human-AI dialogue journaling, and reflexivity with AI-human co-reflection records.

Taken together, these proposals treat generative AI as a provocateur that can challenge dominant narratives and extend analysis, provided that researchers document how they used it and remain accountable for the interpretation. Niroj Dahal reflected on the framework: “If I had to choose one, it would be human-AI dialogue journaling. Keep a record of what you asked the tool, what it gave back, and what you changed and why. It costs nothing and needs no technical expertise. It also makes the other criteria possible, because you cannot be accountable for a process you have not documented.”

Methods and research approach

The article is a perspective piece rather than an empirical study. The authors state that it is based on their personal insights, experiences and observations of using and discussing generative AI in qualitative research, supported by a review of relevant literature on generative AI, research quality and the three crises described by Denzin and Lincoln.

The argument moves through each crisis in turn, from the ontological tension between human subjectivity and AI’s apparent objectivity, to the epistemological uncertainty created by opaque algorithms, to the ethical dilemmas of AI acting as a participant in autoethnography or participatory action research. For praxis, the authors recommend collaborative boundary-setting with communities (for example, limiting AI to “scribe” roles), plain-language disclosure in consent processes, and ethical guidelines that uphold autonomy and informed consent.

As a perspective article, it does not test the proposed criteria empirically. The authors call for standardised disclosure protocols for AI use, decolonised training datasets and longitudinal studies of how generative AI affects research practice over time.

Broader implications

For qualitative researchers, the most immediate implication is to start documenting AI use now: keeping a log of prompts and outputs, recording how human judgement shaped or overruled AI suggestions, and describing AI’s role in consent forms and methods sections. These practices map directly onto the adjusted criteria and make AI-assisted work easier for reviewers and readers to evaluate. Niroj Dahal was clear about where human judgement must remain central: “In interpretation, a generative AI tool lacks embodied experience, cultural context, and relationships with participants, yet it writes in an objective voice. It can suggest themes, but deciding what a participant’s story means and whether it rings true to their lived reality has to stay with the researcher and the community. We argue for prioritizing human accountability over AI efficiency.”

For research leaders, supervisors and journal editors, the framework offers a vocabulary for setting expectations, including what an “algorithmic audit” or an “AI-human co-reflection record” might look like in a thesis or manuscript. The call for interdisciplinary collaboration among ethicists, communities and engineers is also an invitation to institutions in the Global South to help shape the tools themselves. Asked for advice, Niroj Dahal said: “Treat generative AI as a provocateur, not an authority. Let it start a draft, then interrogate what it produces. Label what is AI-generated, disclose its role to your participants and readers, and check its outputs against human coding and member-checking. The meta-crisis is not a threat. It is an invitation to bring the critical spirit of qualitative research back to life.”

It is important to be clear about what the article does not claim. It does not evaluate any particular AI tool, and it does not argue that AI should be avoided. Its contribution is conceptual: a structured way to ask whether AI-assisted qualitative research remains credible, transparent and ethically grounded.

About Niroj Dahal

Niroj Dahal, Kathmandu University School of Education

Niroj Dahal holds a PhD in STEAM Education and has built a career focused on innovative teaching methodologies and educational research. Dr. Dahal works in the Department of STEAM Education at Kathmandu University School of Education. Dr. Dahal’s research interests span a wide range, including ICT in education, qualitative research, transformative research, mathematics education, STEAM education, and the use of generative AI in research and teaching. Dr. Dahal was honored with the Science and Technology Youth Award by the Nepal Academy of Science and Technology (NAST) in December 2025 for his significant contributions to STEAM education and research.

About the authors

  • Niroj Dahal (Kathmandu University School of Education, Lalitpur, Nepal)
  • Md. Kamrul Hasan (Department of English, United International University, Dhaka, Bangladesh)
  • Amine Ounissi (Centre for Foundation and Continuing Education, University Malaysia Terengganu, Malaysia)
  • Md. Nurul Haque (Department of English and Modern Languages, International University of Business Agriculture and Technology, Dhaka, Bangladesh)
  • Hiralal Kapar (Kathmandu University School of Education, Lalitpur, Nepal)

Read the article

Read the full article, “Navigating the meta-crisis of generativity: adapting qualitative research quality criteria in the era of generative AI,” published open access in Frontiers in Research Metrics and Analytics in 2025.

Full citation

Dahal, N., Hasan, M. K., Ounissi, A., Haque, M. N., & Kapar, H. (2025). Navigating the meta-crisis of generativity: Adapting qualitative research quality criteria in the era of generative AI. Frontiers in Research Metrics and Analytics, 10, 1685968. doi.org/10.3389/frma.2025.1685968

View the publication

We are grateful to Dr. Dahal for sharing this perspective and for a contribution to the responsible, transparent use of AI in qualitative research.

Research, accuracy and responsible AI

How Evidano's output has been evaluated, how its models were built, how your data is handled, and how to disclose or cite AI use in your own work.

  • Resource

    Human vs AI: validated accuracy benchmarks

    Four head-to-head comparisons against expert human analysis — 371 interview transcripts with Arizona State and Penn State, 298 evaluation reports, 224 open-ended survey responses with UT Austin and Wichita State, and a 160-response manual-coding review.

  • Resource

    How Evidano’s models were trained

    Model provenance for research-ethics review: a provenance-tiered curated corpus, and customer uploads, prompts and outputs excluded from training entirely.

  • Resource

    Data security for qualitative research

    Encryption in transit and at rest, customer-controlled deletion, regional storage, and SOC 2 Type 2, GDPR, HIPAA-ready and CCPA/CPRA compliance.

  • Resource

    How to cite and disclose Evidano

    Citation formats in APA, MLA, Chicago, Harvard and Vancouver, methods-section wording per workflow, and an AI-policy review of 41 journals across 8 disciplines.