Short answer: There are three main types of thematic analysis. Coding reliability TA uses a fixed codebook and several coders whose agreement is measured. Codebook TA (template, framework analysis) uses a structured codebook but treats coding as interpretive. Reflexive TA treats themes as built by a single researcher's engagement with the data, with no reliability scores. Pick the type that matches your epistemology, then report it by name.
Last updated: November 5, 2026
\n\n\"We used thematic analysis\" is one of the most common sentences in qualitative methods sections, and one of the least informative. Since 2019, Virginia Braun and Victoria Clarke have argued that thematic analysis is not one method but a family, and that most quality problems they see in published papers come from mixing its members. This guide explains the types of thematic analysis, how they differ, how to choose one, and what changes when AI tools do part of the coding.
\n\nThe three types of thematic analysis are coding reliability TA, codebook TA and reflexive TA. They share a goal, finding patterns of meaning across a dataset, but sit on different assumptions about whether meaning is discovered or constructed. Braun and Clarke set out this typology in their 2019 and 2021 work, including \"One size fits all? What counts as quality practice in (reflexive) thematic analysis?\" (Qualitative Research in Psychology, 18(3), 328–352).
\n\nCoding reliability TA treats themes as patterns that exist in the data and can be found accurately. A codebook is built early, often from theory, and two or more coders apply it independently. Agreement is measured with statistics such as Cohen's kappa or Krippendorff's alpha, and disagreements are resolved to raise reliability. Boyatzis (1998) and Guest, MacQueen and Namey's Applied Thematic Analysis (2012) are the usual references. Themes here are often closer to topic summaries (\"participants talked about cost\") than to interpretive claims.
\n\nCodebook TA keeps a structured codebook but drops the idea that agreement proves accuracy. Coding is seen as interpretive, and the codebook is a practical tool for teams, deadlines and pre-set questions. Template analysis (King) and framework analysis (Gale et al., 2013, BMC Medical Research Methodology) belong here. Themes are often developed partly before full coding, and the codebook is revised as analysis goes on.
\n\nReflexive TA is the approach most people mean when they cite Braun and Clarke (2006). Themes are \"generated\" by the researcher, not \"found\" in the data. Coding is open and flexible, usually done by one person or a team aiming for richer interpretation rather than consensus. Researcher subjectivity is a resource, documented through reflexivity, and there is no codebook or reliability score. A theme is a pattern of shared meaning organized around a central concept, not a bucket of everything said about a topic.
\n\n| Dimension | Coding reliability TA | Codebook TA | Reflexive TA |
|---|---|---|---|
| View of meaning | Discovered; closer to (post)positivist | Interpretive, pragmatic | Constructed; fully qualitative |
| Codebook | Fixed early, applied by all coders | Structured, revised during analysis | None; codes evolve freely |
| Coders | Two or more, independent | Often a team | Often one researcher |
| Intercoder agreement | Core quality evidence | Optional, used for consistency | Not meaningful as quality evidence |
| Typical theme | Topic summary | Topic or interpretive theme | Shared meaning around a central idea |
| Typical references | Boyatzis 1998; Guest et al. 2012 | King (template); Gale et al. 2013 (framework) | Braun & Clarke 2006, 2021 |
| Good fit | Large applied studies, deductive questions, audit needs | Policy, health services, multi-site teams | Theory-building, identity and experience research |
Choose by the claim you want to make, not by the tool you have. If you want to say \"this is what is in the data, and independent coders agree,\" you need coding reliability TA. If you need structure for a team or pre-set policy questions but accept that coding is interpretive, use codebook TA. If you want to offer an interpretation that is explicitly yours, use reflexive TA.
\n\nAnswer each question with yes or no. The pattern points to a type.
\nMixed answers are common. If questions 1 and 4 both come out yes, you have a coherence problem to settle before analysis, not after reviewers find it.
\n\nBraun and Clarke's 2021 paper lists recurring problems in published studies. The ones that matter most for choosing a type:
\nFor a worked example of the six phases in reflexive TA, see Byrne (2022) in Quality & Quantity.
\n\nAI fits codebook and coding reliability TA better than reflexive TA. When a codebook exists, a language model can apply it across hundreds of transcripts, and you can check its output against human coders with the same agreement statistics you would use for a second person. Our coverage of the TACT AI-council study describes one such design, with blind human coding and Krippendorff's alpha per code.
\n\nReflexive TA is harder. Its quality depends on a researcher's sustained engagement and reflexivity, and a study we covered on whether LLMs can do reflexive thematic analysis found fabricated quotes among the risks. In reflexive TA, use AI for transcription, retrieval, quote checking and challenging your reading, and keep theme generation and naming yourself. Report exactly what the tool did.
\n\n| Task | Coding reliability TA | Codebook TA | Reflexive TA |
|---|---|---|---|
| Drafting a codebook | Acceptable, then fixed by humans | Acceptable, revised by team | Not applicable |
| Applying codes at scale | Good fit, check agreement | Good fit, spot-check | Support only |
| Generating themes | Possible with review | Possible with review | Keep with the researcher |
| Finding and verifying quotes | Good fit | Good fit | Good fit |
Qualitati is a European qualitative research platform for universities and research companies, with GDPR and data privacy as core priorities. Two of its tools map onto the types above. QDA Workspace supports inductive and deductive coding and codebook generation, which suits codebook and coding reliability workflows where you want a human to review every AI-applied code. ThemeLens maps codes to research questions across up to 100 transcripts and anchors themes in participant quotes, so you can check each claim against its source. For reflexive TA, treat both as assistants and keep interpretive decisions in your own hands.
\n\nWho this is for: PhD students writing a methods chapter, labs designing a team coding process, and research firms that need to defend an analysis to clients. When not to use thematic analysis at all: when your question is about language use (consider discourse analysis), lived experience in depth with very small samples (consider IPA), or building a process theory from theoretical sampling (consider grounded theory).
\n\nCoding reliability TA, codebook TA and reflexive TA, as described by Braun and Clarke. They differ in whether a codebook is used, how many coders work, and whether intercoder agreement counts as quality evidence.
\n\nYes. Braun and Clarke later named their 2006 approach reflexive thematic analysis to separate it from codebook and coding reliability versions that cite the same paper.
\n\nNo. Reflexive TA treats coding as interpretive, so agreement statistics do not measure its quality. Report reflexivity, analytic decisions and a clear audit trail instead.
\n\nBraun and Clarke place framework analysis, like template analysis, within codebook thematic analysis: a structured matrix is used, but coding is treated as interpretive.
\n\nAI can apply a codebook and retrieve evidence at scale, which suits codebook and coding reliability TA. In reflexive TA, theme generation should stay with the researcher, and AI use should be reported.
\n\nThe types of thematic analysis are not interchangeable. Decide whether you are discovering patterns, organizing them for a team, or interpreting them yourself, then name the type, cite the right sources and report the matching quality evidence. If you want to try an AI-assisted codebook workflow with human review, start free with 30 monthly credits, no credit card required, or view transparent pricing. For the six phases in detail, read our thematic analysis guide.
\n\nLast updated: November 5, 2026. Written by the Qualitati Research Team. Sources are cited inline; check the original papers before quoting them in your own work.
\n","related":[{"slug":"multi-llm-thematic-analysis-2026","title":"Multi-LLM Thematic Analysis: Do AI Models Agree? (2026)","description":"Multi-LLM thematic analysis runs several AI models on the same transcripts and keeps only the themes they agree on. See the 2026 reliability evidence and when to use it.","category":"Thematic Analysis","date":"2026-06-11","readTime":"12 min read","author":"Qualitati Research Team"}]}