Reflexive thematic analysis guide
Software for Reflexive Thematic Analysis (Braun and Clarke): A 10-Point Checklist
Most guides to qualitative analysis software compare features: coding speed, price, how many file types it opens. If you are doing Braun and Clarke's reflexive thematic analysis, those are the wrong questions. The method assumes the researcher's interpretation is the analysis, so a good tool is one that keeps it that way. This page gives you ten things to check in any software, AI-assisted or not, a five-minute way to test them, and an honest score for our own tool against the same list.
What should software for reflexive thematic analysis do?
It should keep you as the author, keep themes provisional, trace every claim to the data, record what any AI did, and leave room for your own reflexivity. Concretely, check that:
- You stay the author.
- Themes stay provisional.
- Every claim traces to the data.
- Disagreement is shown, not smoothed.
- There is a record of what the AI did.
- There is room for reflexivity.
- Themes are claims, not counts.
- It tells you what it left out.
- You know where your data goes.
- You can take your work with you.
Braun and Clarke have not endorsed any software, and this page is not their endorsement of ours.
First: should AI be in your analysis at all?
This comes before any feature comparison, because the people who wrote the method disagree with using generative AI in it. In 2025 Tanisha Jowsey, Virginia Braun, Victoria Clarke, Deborah Lupton and Michelle Fine published an open letter rejecting generative AI for reflexive qualitative research. Their argument is that reflexive analysis needs a positioned human researcher at every phase, including coding. Other researchers replied that critical, researcher-led use can be compatible with reflexive work (for example Friese and colleagues, 2026). Our guide to AI qualitative coding sets out both sides.
That leaves three defensible positions, and you should pick yours before you pick a tool:
- No AI. You code and build themes yourself. Use any software for organisation only, with its AI features off, or work on paper, Word or Excel. If this is your position or your supervisor's, the rest of this page is about the non-AI features to look for (items 1, 6, 7, 9 and 10 still apply).
- AI as a bounded aid. You read the data, do the interpretive work, and use AI to organise or to give you a draft to react to. The checklist below is written for this position.
- AI as a first pass. The AI codes and groups, and you review everything. This needs the most checking and the most honest reporting.
The 10-point checklist
Each point comes with why it matters in reflexive thematic analysis, a quick way to test it, and the red flag to watch for. The method itself is covered in our guide to Braun and Clarke's thematic analysis.
Check 1
You stay the author
Why it matters: Reflexive thematic analysis treats the researcher's interpretation as the analysis itself, not as noise to be removed (Braun and Clarke, 2019, 2021).
Test it: Rename a theme and move a code to another theme. Does the change stick, and is it recorded next to what the software proposed?
Red flag: Output presented as final, or your edits lost when you run it again.
Check 2
Themes stay provisional
Why it matters: Braun and Clarke say themes are constructed by the researcher, not found in the data, and developed over several passes.
Test it: Can you ask for the themes to be rebuilt from a different angle, and does the earlier version survive?
Red flag: Language about "discovering" or "extracting" themes, or a one-shot result.
Check 3
Every claim traces to the data
Why it matters: A theme you cannot trace to passages cannot be defended in a viva or a peer review.
Test it: Pick three quotes the software gives you and search for them in your own transcript. Click a code: does it take you to the passage?
Red flag: Quotes that are paraphrased, trimmed or absent from the source. General chatbots have been found to alter quotes (see our guide on whether ChatGPT can do thematic analysis).
Check 4
Disagreement is shown, not smoothed
Why it matters: Accounts that differ, or cut against a theme, often show you where the theme stops being true.
Test it: Does each theme say where participants differ, or where a case does not fit?
Red flag: Every theme described as shared by "most participants", with nothing that cuts against it.
Check 5
There is a record of what the AI did
Why it matters: You will have to describe the tool's role in your methods section, and you cannot describe what you cannot see.
Test it: Look for the model name and version, the settings, what the software discarded and what you changed.
Red flag: "Proprietary AI" with no record of what it did to your data.
Check 6
There is room for reflexivity
Why it matters: A positionality statement and reflexive notes are part of the method, not an add-on.
Test it: Is there somewhere to write a positionality statement before you start, and a place for a note on each theme? Can both be exported?
Red flag: No place for researcher notes, or notes you cannot take out of the app.
Check 7
Themes are claims, not counts
Why it matters: In reflexive thematic analysis a theme earns its place by what it says about the data, not by how many people mentioned it.
Test it: Are theme names statements, or topic labels like "Workload"? Is the main result a ranking by frequency?
Red flag: Topic headings and bar charts of mentions as the headline output.
Check 8
It tells you what it left out
Why it matters: Any first pass skips passages. You need to see them to decide whether the gap matters.
Test it: Is there a coverage view, or a list of passages that no code touched?
Red flag: No way to see what was not coded.
Check 9
You know where your data goes
Why it matters: Participants' words are covered by your ethics approval, and your committee will ask.
Test it: Does the vendor state in writing whether data is stored, whether it is used to train models, and which AI provider processes it?
Red flag: Vague privacy language, or certifications you cannot verify.
Check 10
You can take your work with you
Why it matters: A supervisor, a co-author or a later stage of the project may need another tool.
Test it: Can you export to Word and a spreadsheet, and to a REFI-QDA (.qdpx) project file, the exchange format that NVivo, MAXQDA and ATLAS.ti use to import from other software?
Red flag: Results that live only inside the vendor's app.
Check 7 is the one most often missed. A tool that ranks themes by how many participants mentioned them is quietly teaching you a different method, one where importance is a tally. In reflexive work it is a judgement about what a pattern says in relation to your research question. Our guide to codes versus themes explains why a topic label is not a theme.
A five-minute test for any tool
You do not need to read a vendor's documentation to find out whether it passes. Run this on a small piece of data you know well.
- Use a short piece of real data. Two or three pages of a transcript you already know well, so you can judge the output.
- Check three quotes. Search for each one in your transcript. Word for word, or close enough to fool you?
- Change something. Rename a theme, move a code, ask for a rebuild. Did your change survive, and is it logged?
- Hunt for the record. Find the model name, the settings and the list of what was dropped. If you cannot find them, you cannot report them.
- Look for reflexivity. Find the place for a positionality statement and a note on a theme. Try exporting both.
- Check the exit. Export everything. Open it somewhere else. Is it usable?
Most failures show up at step 2 or step 4. If a tool cannot show you where a quote came from, or what it did to your data, you will have nothing to put in your methods section.
Where software fits in the six phases
A tool can help with the mechanical parts of each phase. What it should not do is the interpretive part. Phase names below are from the 2006 paper; Braun and Clarke have relabelled some phases in later work.
| Phase | You | Software can | Software should not |
|---|---|---|---|
| 1. Familiarising with the data | Read the data, note first impressions | Transcribe, format, split one file into participants | Summarise in place of your reading |
| 2. Generating initial codes | Code, or review suggested codes against their passages | Propose first-pass codes, show the source quote for each | Decide which codes matter |
| 3. Searching for themes | Cluster codes, write each theme's central idea | Show a draft grouping for you to compare with yours | Hand you a grouping as the finding |
| 4. Reviewing themes | Test themes against the whole dataset, revise | Surface divergent and uncoded passages | Approve its own themes |
| 5. Defining and naming themes | Write the definitions and names | Suggest wording to react to | Name them without your review |
| 6. Producing the report | Write up your interpretation | Export tables, quotes and the audit record | Write your interpretation |
Phase 3 is where most people stall. Our guide to going from codes to themes shows the merge, cluster and test steps on a worked example, and reflexivity in thematic analysis covers the positionality statement.
What the established tools say about their own AI
This is how each vendor described its AI features on its own site on 29 October 2026. Features change, so check the current page, and test each one against the checklist yourself rather than taking the description as proof.
| Tool | What its site says about its AI |
|---|---|
| NVivo AI Cloud (Lumivero) | Click-through citations to the source, human approval before the AI writes anything, suggested topics you confirm or reject (or your own framework), an exportable audit trail, and a "Devil's Advocate" mode that surfaces disconfirming evidence. |
| ATLAS.ti | "Intentional AI Coding", which codes against research goals you state, plus a conversational AI assistant. |
| MAXQDA | AI Assist, and a local AI option in MAXQDA 27 for analysis without sending data to a cloud service. |
Several of these map neatly onto the checklist: citations (check 3), human approval and an audit trail (checks 1 and 5), and disconfirming evidence (check 4). None of these descriptions, and none of ours, is an endorsement from Braun and Clarke. If you are comparing tools in more depth, see our guides to NVivo and MAXQDA, and our thematic analysis software comparison.
How thematicanalysis.ai scores on its own list
We built thematicanalysis.ai to run reflexive thematic analysis by default, so it should hold up against this checklist. Here is our own marking. This reflects how the tool worked when this guide was published.
| Check | Score | Why |
|---|---|---|
| You stay the author | Meets, with a caveat | You can rename, merge, move and remove everything, add your own codes to any passage in the source reader, and load your own codebook so the coding uses your codes (hybrid or strict). Your changes are logged in the audit trail beside what the AI proposed. The caveat is that by default the AI writes the first draft, which is what critics of AI in reflexive work object to. |
| Themes stay provisional | Meets | An "Ask for different themes" box rebuilds them from the same codes following what you say, and each rebuild is recorded. |
| Every claim traces to the data | Meets | Each code carries a verbatim quote checked against your uploaded text. Click a code to see the passage highlighted. Quotation marks in the written theme narratives are checked too, and sentences that claim more than the evidence supports are removed, with each removal recorded. |
| Disagreement is shown | Meets | Every theme records where sources differ, sorted into six kinds (emphasis, scope, degree, mechanism, method and contradiction), each flagged as limiting the theme's claim or not, with references checked against the text. Where sources agree, the theme still records what was compared. It is part of every theme rather than a separate mode. |
| A record of what the AI did | Meets | The audit trail covers the model and prompt version, settings, a hash of each source, what verification dropped, merges and your changes. It describes itself as a faithful account for your methods chapter, not a tamper-proof record. The full appendix is part of paid reports; free reports show it locked. |
| Room for reflexivity | Meets | An optional positionality statement written before the run, a reflexivity memo on each theme, and questions built from what that run did. Both go into the Word report. The tool cannot do the reflecting for you. |
| Claims, not counts | Meets | The theme-building step is instructed to produce patterns of shared meaning organised around one central concept, not topic buckets, and each theme records what it is not. The participants behind each theme are listed, so counts are visible. What they mean is yours to decide. |
| Tells you what it left out | Meets | Every report has a coverage table. The uncoded passages themselves are listed in paid reports. |
| Where your data goes | Meets | We keep no copy, the text is processed by OpenAI's API and not used to train models, and there is no account. We make no certification claims, and the privacy policy sets out exactly what is processed. |
| Take your work with you | Meets | Word, Excel, CSV and a REFI-QDA (.qdpx) project file. |
Don't take our word for it. Run the test from the section above on your own data: paste a few interviews below, keep the speaker labels in, and check the quotes, the record and the memo space against the list.
Describing your software in your methods section
Whatever you choose, say what it did and what you did. A reader should be able to see where the software stopped and your interpretation began. For example:
I followed Braun and Clarke's reflexive thematic analysis. I read all transcripts in full before any coding. A software tool proposed initial codes with supporting quotes, which I checked against the transcripts, revising or removing those that did not hold. I grouped codes into themes myself, using the software's draft grouping only as a comparison, and wrote each theme's central idea and definition. My positionality statement and reflexive memos are in Appendix B.
Adapt it to what you actually did, because your methods section has to be true. Our guide to reporting AI-assisted thematic analysis has more wording, and the dissertation guide shows where it sits in a thesis.
Frequently asked questions
Is there software designed for reflexive thematic analysis?
Reflexive thematic analysis is a method, not a product. Most qualitative analysis packages let you do the six phases by hand, and what differs between tools is how well they keep the interpretation yours. We know of no software endorsed by Braun and Clarke, so treat any claim of their endorsement with suspicion and ask for the source.
Can AI do reflexive thematic analysis?
It is contested. In a 2025 open letter, Jowsey, Braun, Clarke, Lupton and Fine rejected generative AI for reflexive qualitative research, arguing that it needs a positioned human researcher. Others, including Friese and colleagues (2026), argue that researcher-led, critical use is compatible with reflexive work. Which position you take affects your methodology, so agree it with your supervisor before you start.
Do NVivo, MAXQDA and ATLAS.ti support reflexive thematic analysis?
They all support the manual work of coding, memoing and grouping, and each now has AI features as well. Our guides to using NVivo and MAXQDA for thematic analysis show how the six phases map onto their tools. The checklist above applies to their AI features as much as to ours.
How do I describe software in my methods section?
Say which tool you used, for which phase, what it produced, and what you did with it: which codes you changed, how you built the themes, and how you checked quotes against the data. Name the model or version if the tool reports one. Our guide to reporting AI-assisted thematic analysis has example wording.
What if my supervisor does not accept AI in the analysis?
Follow their guidance. You can still use software without AI features, or code on paper or in Word or Excel. NVivo, MAXQDA and ATLAS.ti all work for manual coding, and QualCoder and Taguette are free open-source options.
Do I need software for reflexive thematic analysis at all?
No. Many researchers work with paper, highlighters, Word or Excel, and Braun and Clarke's method does not require software. Software mainly helps with organisation and retrieval when the dataset is large.
How this guide was made
Written by the thematicanalysis.ai team, who also make one of the tools discussed, so read our scoring of ourselves with that in mind. We used AI to help research and draft it, then checked the method claims against the sources below and edited the text. The descriptions of NVivo, ATLAS.ti and MAXQDA come from their own websites as read on 29 October 2026 and may have changed. The description of our tool reflects how it worked when this guide was published.
References
- Braun, V. and Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), pp. 77–101. doi:10.1191/1478088706qp063oa
- Braun, V. and Clarke, V. (2019). Reflecting on reflexive thematic analysis. Qualitative Research in Sport, Exercise and Health, 11(4), pp. 589–597. doi:10.1080/2159676X.2019.1628806
- Braun, V. and Clarke, V. (2021). One size fits all? What counts as quality practice in (reflexive) thematic analysis? Qualitative Research in Psychology, 18(3), pp. 328–352. doi:10.1080/14780887.2020.1769238
- Braun, V. and Clarke, V. Thematic analysis: frequently asked questions. University of Auckland. thematicanalysis.net/faqs
- Friese, S., Nguyen-Trung, K., Powell, S. and Morgan, D.L. (2026). Beyond binary positions: making space for critical and reflexive GenAI integration in qualitative research. Qualitative Inquiry. doi:10.1177/10778004261429393
- Jowsey, T., Braun, V., Clarke, V., Lupton, D. and Fine, M. (2025). We reject the use of generative artificial intelligence for reflexive qualitative research. Qualitative Inquiry. doi:10.1177/10778004251401851