03 · What You Need to Know
Why Delegating Interpretation Creates a Methodological Problem
What Makes Reflexive Thematic Analysis Reflexive?
Reflexive thematic analysis is an interpretive approach to examining patterns of shared meaning across qualitative data. Braun and Clarke developed the approach through their foundational 2006 publication and subsequently clarified its philosophical assumptions and analytical practices.
A defining feature is that the researcher does not simply discover themes that already exist independently within the dataset. Rather, themes are developed through active engagement with the data, informed by the researcher's theoretical orientation, questions, assumptions, and interpretive decisions.
This does not mean that the researcher can impose any interpretation they prefer. Analytical claims must remain grounded in the data and coherent with the research question and theoretical framework.
Reflexivity involves critically examining how the researcher's position and assumptions shape the analytical process. Researcher subjectivity is therefore not necessarily treated as contamination that must be eliminated. It is recognized as part of knowledge production and requires thoughtful engagement.
These commitments distinguish reflexive thematic analysis from approaches that emphasize standardized classification and agreement among coders.
Why Following the Six Phases Is Not Sufficient
Braun and Clarke describe six phases of reflexive thematic analysis: familiarization, coding, generating initial themes, developing and reviewing themes, refining and naming themes, and writing up.
These phases are recursive analytical activities rather than a rigid algorithm.
A generative AI system can produce text corresponding to each phase. For example, it may summarize transcripts under familiarization, generate code labels, cluster those codes into candidate themes, and write a thematic narrative.
However, producing the expected outputs does not establish that the analytical process occurred in the methodological sense intended by reflexive thematic analysis.
Reproducing the procedural outputs
Generating summaries, codes, theme labels, and narrative explanations organized according to the six phases.
Engaging in reflexive thematic analysis
Developing patterns of meaning through situated researcher engagement, theoretical interpretation, and critical reflection on analytical decisions.
The difference is methodological rather than cosmetic. A report can resemble reflexive thematic analysis in structure while relying on analytical procedures that are inconsistent with its assumptions.
What Did Braun and Clarke Say About Generative AI?
The question is no longer confined to speculation about what the original authors might think.
In December 2025, a statement published online in Qualitative Inquiry, titled We Reject the Use of Generative Artificial Intelligence for Reflexive Qualitative Research, explicitly opposed generative AI use in reflexive qualitative approaches.
The statement was authored by Tanisha Jowsey, Virginia Braun, Victoria Clarke, Deborah Lupton, Michelle Fine, and numerous additional signatories. It argues that approaches such as reflexive thematic analysis require a subjective, positioned, and reflexive human researcher and that generative AI is not methodologically congruent with these practices.
The authors also raise concerns relating to social and environmental justice.
This is a substantive methodological position, not a universal prohibition governing every form of qualitative research. It is nevertheless particularly consequential for researchers who explicitly claim to follow Braun and Clarke's reflexive thematic analysis.
A researcher who delegates interpretation to AI while citing Braun and Clarke should recognize that the approach's developers have publicly rejected that use. Describing the workflow as methodologically unproblematic would misrepresent the current debate.
Why AI-Generated Interpretation Is Not Equivalent to Researcher Interpretation
Generative AI can construct explanations of participants' statements and identify possible conceptual relationships. It may produce sophisticated language about power, identity, inequality, or lived experience.
However, generating such language does not establish that the system occupies a situated relationship to the research setting or has engaged in reflexivity as understood within the methodology.
Consider a researcher investigating how university students experience mandatory camera use during online classes.
An AI system might interpret camera requirements as institutional surveillance. Another prompt might lead it to emphasize social presence and belonging. Both interpretations could sound theoretically plausible.
The question is not merely which explanation sounds more convincing. It is how the interpretation was developed, what assumptions shaped it, how the researcher engaged with contradictory accounts, and how the analytical claim relates to the study's theoretical orientation.
A model's generated explanation cannot substitute for that situated analytical responsibility simply because the output is fluent.
Can AI Have Reflexivity in the Methodological Sense?
A language model can generate reflective statements about assumptions, biases, and alternative interpretations. It can even write in the first person, describing supposed changes in its understanding.
But producing reflective language is not equivalent to the human researcher's reflexive examination of their social position, experience, and relationship with participants.
For example, AI might generate the statement:
"I initially interpreted students' reluctance to activate their cameras as disengagement, but my engagement with the transcripts led me to reconsider my assumptions about participation."
If the researcher did not actually undertake that reflection, the statement would fabricate an analytical history.
Even if the model generated a similar statement about its own processing, that would not establish the kind of situated reflexivity required by the methodological approach.
The distinction between producing reflexive text and exercising reflexivity deserves careful attention when evaluating AI-assisted qualitative research.
Why Positionality Matters to the Interpretation
Positionality concerns the researcher's relationship to the research context, including relevant social, professional, institutional, and experiential circumstances.
A university instructor studying online classroom participation may interpret students' accounts differently from a researcher whose experience lies primarily in student advocacy. Neither perspective automatically guarantees a better interpretation.
What matters is how the researcher recognizes and critically examines the influence of those positions.
Generative AI can describe possible positional influences, but it does not occupy the researcher's actual relationship with the participants or institution.
Consequently, delegating interpretation to a model raises questions about whose standpoint is shaping the analysis and how that standpoint can be examined.
The issue of whether AI can have positionality should not be resolved simply by asking the model to generate a positionality statement.
Why Coding Reliability Is Not the Same as Reflexive Analytical Quality
Researchers sometimes justify AI-assisted reflexive thematic analysis by reporting high agreement between AI-generated and human-generated codes.
Agreement may be useful when evaluating a structured classification task. However, Braun and Clarke explicitly distinguish reflexive thematic analysis from coding reliability approaches.
In reflexive thematic analysis, coding is an interpretive activity. Different researchers may develop different, defensible readings of the same material. Agreement is therefore not treated as the universal measure of analytical quality.
A model that reproduces a human researcher's code labels with high accuracy has demonstrated something about classification similarity under the evaluation procedure. It has not necessarily demonstrated reflexive engagement or interpretive adequacy.
Likewise, a model that generates the same theme names as a human analyst has not necessarily followed the same analytical process.
Researchers should avoid applying quality criteria from one thematic analysis approach to another without considering their methodological assumptions.
What Does Empirical Research Show About AI and Reflexive Thematic Analysis?
Empirical studies have explored the use of large language models to generate themes from qualitative interviews and compared their outputs with human analyses.
Some evaluations report similarities between AI-generated and researcher-generated themes. Others identify difficulties involving contextual meaning, interpretation, fabricated quotations, and superficial theme development.
A 2026 study titled The Role of Generative Artificial Intelligence in the Analysis of Qualitative Data Compared With Human-Led Analysis examined AI-generated themes from a secondary dataset of 23 interviews that had previously undergone human-led analysis. The study provides an example of researchers experimentally applying generative AI to a workflow described in relation to reflexive thematic analysis.
Such comparisons can inform judgments about the usefulness and limitations of AI-generated outputs. However, empirical similarity does not independently resolve the methodological question of whether delegated interpretation constitutes reflexive thematic analysis.
This distinction is central to the debate: a system may approximate the products of an analytical method without satisfying the philosophical commitments through which that method defines knowledge production.
Why a Human Review at the End Does Not Automatically Solve the Problem
Suppose AI generates all codes, develops the themes, and writes the interpretation. The researcher then reviews the output and confirms that the quotations are accurate.
This verification is valuable, but it does not necessarily establish that the researcher engaged in the interpretive process required by reflexive thematic analysis.
Checking whether an output is plausible differs from developing an interpretation through sustained engagement with the data.
The issue is especially consequential when the researcher's role becomes limited to approving or editing themes generated elsewhere.
Researchers should therefore distinguish between human oversight of an AI-led workflow and human-led reflexive interpretation.
Are All Forms of AI Assistance Equally Problematic?
No. It is useful to distinguish analytical delegation from other forms of technological assistance, although the boundaries are not always straightforward.
| AI-related activity |
Nature of involvement |
Methodological consideration |
| Formatting transcript files |
Administrative or technical processing |
Check accuracy, confidentiality, and whether meaningful transcript features are preserved. |
| Summarizing interviews |
Selective representation of participant accounts |
May influence familiarization and omit analytically important details. |
| Suggesting initial codes |
Participation in analytical interpretation |
Raises questions about whose interpretive choices shape the analysis. |
| Generating themes |
Developing proposed patterns of meaning |
Directly implicates the interpretive work central to reflexive thematic analysis. |
| Writing reflexive statements |
Representing assumptions and analytical experiences |
Cannot substitute for authentic researcher reflection. |
| Editing a researcher-written manuscript |
Language and presentation assistance |
Must preserve the researcher's interpretation and comply with applicable policies. |
This table distinguishes activities for methodological examination. It should not be interpreted as an endorsement of generative AI use within reflexive thematic analysis. The 2025 statement by Braun, Clarke, and colleagues takes a more categorical position against generative AI in reflexive qualitative research.
Researchers who follow that position may choose not to use generative AI in the analytical process at all.
Could Another Qualitative Approach Better Fit an AI-Led Workflow?
Potentially. Some structured qualitative approaches place greater emphasis on predefined coding frameworks, consistent category application, or systematic comparison across cases.
For example, a study using deductive qualitative content analysis might define categories in advance and evaluate how accurately AI applies those categories.
That workflow can be assessed using criteria appropriate to structured coding, including classification errors and consistency.
However, changing the methodological label does not automatically make the analysis valid. The researcher must still justify the approach, verify the evidence, and explain how the findings were developed.
The key question is whether the analytical procedure fits the research question and epistemological assumptions. This connects to the broader issue of how AI-assisted coding relates to qualitative epistemology.
Can Researchers Simply Redefine Reflexive Thematic Analysis to Include AI?
Researchers may develop and defend new methodological approaches, including approaches that integrate generative AI into qualitative interpretation.
However, a modified method should not automatically be represented as equivalent to Braun and Clarke's reflexive thematic analysis.
If AI performs substantial interpretive work, the researcher should explain how the method differs from established reflexive thematic analysis, what assumptions justify the modification, and how analytical quality is evaluated.
Methodological innovation requires conceptual justification, not merely the addition of a new technology to an existing procedure.
Researchers should also avoid describing an AI-led workflow as reflexive simply because the prompt contains words such as "critical," "reflective," or "interpretive."
04 · A Practical Example
When an AI-Generated Thematic Analysis Is Presented as Reflexive
Hypothetical Example
Understanding Students' Experiences of Camera Requirements
A researcher conducts 20 semi-structured interviews about university students' experiences of mandatory camera use during online classes.
The researcher states that the study will use Braun and Clarke's reflexive thematic analysis but asks a generative AI system to perform the coding and develop the themes.
For illustration, the following participant statements are invented:
-
Participant A: "I keep my camera off because I don't want everyone to see our house."
-
Participant B: "Our instructor thinks we're not paying attention if our cameras are off."
-
Participant C: "I feel more connected to my classmates when our cameras are on."
-
Participant D: "Sometimes my internet connection cannot support video."
What AI Might Produce
The model generates three themes:
- Privacy and Socioeconomic Vulnerability
- Institutional Surveillance and Student Compliance
- Social Presence and Classroom Belonging
It then writes a polished narrative explaining how camera requirements reproduce inequalities while simultaneously supporting social connection.
Some of these ideas may be worth investigating. However, the model has introduced claims about socioeconomic vulnerability and institutional surveillance that require more evidence than the short excerpts provide.
More fundamentally, the model has constructed the analytical framework rather than the researcher developing it through reflexive engagement with the data.
Why Human Approval Does Not Make the Process Reflexive
Suppose the researcher checks the quotations, renames one theme, and approves the final interpretation.
The findings may become more accurate through that review. Nevertheless, the researcher's involvement remains primarily evaluative rather than constitutive of the interpretation.
Calling the process reflexive thematic analysis would require an explanation of how the analytical practice remains consistent with the method's commitments to researcher subjectivity and reflexivity.
Under the position articulated by Braun, Clarke, and colleagues in 2025, the use of generative AI for such reflexive qualitative analysis would not be methodologically congruent.
What a Methodologically Consistent Alternative Looks Like
1. Establish the methodological orientation.
Determine whether the research question calls for reflexive thematic analysis or a different approach emphasizing structured coding.
2. Engage directly with the transcripts.
For reflexive thematic analysis, the researcher develops familiarity with the data through reading and critical engagement.
3. Develop codes through researcher interpretation.
Generate and revise codes in relation to the research question, theoretical assumptions, and emerging understanding.
4. Construct and refine themes.
Develop patterns of shared meaning, examine their central organizing concepts, and revisit the wider dataset.
5. Document reflexive engagement.
Record genuine analytical questions, assumptions, uncertainties, and changes in interpretation.
6. Report the actual analytical process.
Describe the methodology and any technological assistance accurately, without attributing unperformed activities to the researcher.
If the researcher instead intends to delegate substantial classification or interpretation to AI, a different methodological design may be more defensible. That design would require its own justification and evaluation rather than simply adopting the reflexive thematic analysis label.
06 · What This Means for You
How to Decide Whether AI Delegation Fits Your Qualitative Methodology
The decision should begin with the methodological purpose of the study, not with the capabilities of the AI tool.
If you intend to follow Braun and Clarke's reflexive thematic analysis, the interpretive role of the researcher is central. Delegating that role to AI creates a substantive methodological inconsistency, not merely a documentation problem.
Choose the Analytical Approach Before Choosing the AI Workflow
If you are committed to Braun and Clarke's reflexive thematic analysis
Conduct the interpretive analysis through researcher engagement and take seriously the authors' explicit position against generative AI use in reflexive qualitative research.
If your objective is systematic application of predefined categories
Consider a structured coding approach that permits explicit classification rules and appropriate validation.
If AI has already generated your codes and themes
Do not retroactively describe the process as researcher-led reflexive analysis. Reconsider the methodological designation or undertake an appropriate analysis consistent with the intended method.
If you are proposing a new AI-integrated interpretive methodology
Explain its philosophical assumptions, analytical procedures, limitations, and differences from established reflexive thematic analysis.
Do Not Use Prompting to Manufacture Reflexivity
Instructions such as "act as a reflexive qualitative researcher" or "critically reflect on your positionality" may produce persuasive prose. They do not establish the situated human engagement required by the methodology.
A prompt cannot resolve a methodological incompatibility simply by instructing the model to behave as though it possesses the researcher's experience or social position.
Researchers should avoid including generated first-person reflections that describe analytical activities, assumptions, or changes in understanding that did not actually occur.
Distinguish Methodological Adaptation From Methodological Mislabeling
Researchers may legitimately investigate new AI-assisted approaches to qualitative analysis. Such work can be valuable when the research question and evaluation procedures are clearly specified.
However, the use of an established methodological label carries obligations to represent that approach accurately.
If AI substantially determines the codes, themes, and interpretations, the study should explain that arrangement and justify the resulting method. It should not imply that researcher-led reflexive interpretation occurred merely because the final report follows a familiar structure.
Document the Actual Role of AI
When AI is used in qualitative research, document what material was processed, what analytical tasks the system performed, and how the researcher evaluated its outputs.
Such documentation does not automatically establish methodological congruence. Transparency is necessary for evaluating the procedure, but disclosure cannot transform an incompatible analytical process into a compatible one.
The distinction between accurate reporting and methodological justification is central to reporting AI use in qualitative data analysis.
Protect Participants and Consider Broader Ethical Implications
Interview transcripts may contain sensitive information and contextual details that permit participant identification. Researchers must verify whether AI processing is permitted by consent, ethics approval, institutional policy, and applicable data protection requirements.
Ethical considerations also extend beyond confidentiality. The 2025 statement opposing generative AI in reflexive qualitative research raises social and environmental justice concerns, emphasizing that methodological decisions about AI may have consequences beyond analytical efficiency.
Researchers should therefore evaluate AI use as both a methodological and ethical decision rather than assuming that technological convenience is sufficient justification.
