What are the six phases of thematic analysis?

In Braun and Clarke’s 2022 wording, the phases are familiarising yourself with the dataset; coding; generating initial themes; developing and reviewing themes; refining, defining and naming themes; and writing up. The 2006 article used older names for some of them, such as ‘searching for themes’ instead of ‘generating initial themes’ and ‘producing the report’ instead of ‘writing up’. The process is recursive, so you move back and forth between phases.

How is reflexive thematic analysis different from codebook thematic analysis?

Codebook approaches, such as template and framework analysis, use a structured coding frame and usually develop themes early, often as summaries of what participants said about each topic. Reflexive TA uses no codebook: codes evolve throughout coding, and themes are developed late, as the outcome of coding and interpretation. Both differ from coding reliability approaches, which measure agreement between independent coders.

Do I need a second coder for reflexive thematic analysis?

No. Braun and Clarke argue that coding in reflexive TA is interpretation, so there is no correct coding for a second coder to confirm, and inter-rater reliability scores do not fit the approach. Discussing your coding with your supervisor is still useful, as long as the aim is to reflect on your assumptions and readings rather than to reach agreement.

Reflexive thematic analysis: Braun and Clarke’s six phases explained

Updated 11 October 2026 · 10 min read · By Spire Light

Reflexive thematic analysis is Virginia Braun and Victoria Clarke’s method for developing, analysing and interpreting patterns of shared meaning, called themes, across qualitative data such as interview transcripts. You code the data, generate themes from the codes and write an interpretation of what they say about your research question. The word ‘reflexive’ signals that your own perspective is part of the analysis, not a source of error to be removed.

What is reflexive thematic analysis?

Braun and Clarke first set out their approach in Using thematic analysis in psychology, published in Qualitative Research in Psychology in 2006 (3(2), 77–101). The article argued that thematic analysis is a method in its own right, flexible because it is not tied to one theory. Many studies citing it have since used procedures it never recommended, such as fixed codebooks and agreement scores between coders.

In Reflecting on reflexive thematic analysis (2019), published in Qualitative Research in Sport, Exercise and Health 11(4), they explained why they now call their version reflexive thematic analysis: to set it apart from other approaches with the same label. The fullest account is their book Thematic Analysis: A Practical Guide (2022), which uses revised names for the six phases. The authors write ‘reflexive TA’ rather than ‘RTA’, because the acronym hides the reflexivity at its centre.

Reflexive TA suits most questions about experiences, views, practices and how something is represented. The exception, according to the authors’ website, is questions about language practice. If your dissertation follows reflexive TA, cite the 2019 article or the 2022 book alongside the 2006 paper.

What makes it reflexive?

Braun and Clarke place reflexive TA in what Kidder and Fine (1987) called Big Q qualitative research: qualitative methods within a qualitative paradigm, where meaning is situated. Small q research uses qualitative data but keeps positivist aims such as objectivity. Three things follow for your analysis.

Your subjectivity is a resource. Your background, theoretical commitments and reading shape what you notice. In reflexive TA that is not bias to be controlled but what makes interpretation possible, so you reflect on it, for example in a reflexive journal and a positionality statement.
Themes are generated, not found. As early as 2006, Braun and Clarke described an account of themes ‘emerging’ as ‘a passive account of the process of analysis’ (p. 80). You build themes from codes, which is why the 2022 book calls phase three ‘generating initial themes’ instead of ‘searching for themes’. Write that you developed themes, never that they emerged.
You state your choices. Coding can lean towards the data (inductive) or towards existing theory (deductive), and towards what participants say explicitly (semantic) or the assumptions underneath (latent). Neither end is better, but the choice must fit your research question. The guide to inductive vs deductive coding explains the first pair.

Braun and Clarke’s six phases of thematic analysis

The phases below use the 2022 names. The 2006 names, which many module handbooks still use, are in brackets. Treat the phases as a guide, not a recipe: analysis is recursive, and you will move back and forth, especially between phases three and five.

1
Familiarising yourself with the dataset
(2006: ‘Familiarizing yourself with your data’.) Read every transcript more than once, ideally while listening to the recording, and make notes on each interview and on the dataset as a whole, including your own reactions. Checking automatic transcripts against the audio is part of this phase.
2
Coding
(2006: ‘Generating initial codes’.) Work through the whole dataset and label each segment relevant to your research question with a code, a short label that captures one meaning. An extract can carry several codes. Codes will be renamed, merged and split, so plan for more than one round.
3
Generating initial themes
(2006: ‘Searching for themes’.) Look for clusters of codes that share a central organising concept, the core idea that holds them together. Collate the extracts under each candidate theme and sketch a thematic map. Every theme is still provisional.
4
Developing and reviewing themes
(2006: ‘Reviewing themes’.) Test each candidate theme against its extracts, then against the full dataset. Does it hold together, answer part of your question and stand apart from the others? Themes may be split, combined or discarded.
5
Refining, defining and naming themes
(2006: ‘Defining and naming themes’.) Write a short definition of each theme: its central organising concept and its boundaries. Give it a name that conveys the meaning, perhaps a phrase from the data.
6
Writing up
(2006: ‘Producing the report’.) Writing is part of the analysis. You weave together analytic narrative and data extracts, and keep refining the themes as you write.

Phases one to four send you back to the same extracts again and again, which is quicker when transcripts and codes sit in one place. Check your university’s rules before you upload recordings anywhere: a voice identifies the speaker, so a recording is personal data even without names. Spirelight offers no data processing agreement; your university, not you, would sign one. Transcription runs on EU servers, files are stored with Google Firebase in the US, and AI features run on OpenAI in the US, which your ethics application and participant information sheet must state.

Where your university allows it, you can transcribe your interviews and code them in one project that you can share with your supervisor. The code overview gathers every extract with a given code into one table, which helps in phases three and four. The tool keeps codes and extracts in order; it does not generate themes for you.

Reflexive, codebook and coding reliability approaches

Thematic analysis is a family of methods. Braun and Clarke (2022) sort it into three clusters that differ in their assumptions about knowledge as much as in their procedures.

Aspect	Coding reliability TA	Codebook TA	Reflexive TA
Examples	Boyatzis (1998); Guest, MacQueen and Namey (2012)	Template analysis, framework analysis, matrix analysis	Braun and Clarke (2006, 2019, 2022)
Paradigm	Small q: aims for objective, unbiased coding	Qualitative values, with pragmatic compromises	Big Q: fully qualitative
Codebook	Fixed codebook, often set up early	Structured coding frame: a template, framework or matrix	None: codes evolve throughout coding
When themes are developed	Early, sometimes before coding	Mostly early, then refined	Late, as the outcome of coding
Coders	Several independent coders; agreement measured, for example with Cohen’s kappa	Often a team; agreement is not usually a quality measure	One researcher is enough; discussion serves reflexivity
Researcher subjectivity	A threat to be contained	Acknowledged	A resource for analysis

Codebook approaches suit applied projects with fixed information needs, tight deadlines and teams that share the coding, and their themes are often summaries of answers to each question. Reflexive TA suits a dissertation in which you interpret meaning in depth yourself. Whichever you choose, be consistent. A methodology chapter that cites Braun and Clarke but reports a fixed codebook and a kappa score mixes approaches built on different assumptions.

Themes versus topic summaries

In reflexive TA a theme is a pattern of shared meaning underpinned by a central organising concept. A topic summary collects what participants said about a topic, often one of your interview questions. Its extracts share a topic but not a meaning. Braun and Clarke give ‘Experiences of Y’ and ‘Benefits of X’ as typical names of topic summaries presented as themes.

If each of your themes matches a question you asked, you probably have topic summaries. Try stating the central organising concept in one sentence. ‘Supervision’ is a topic. ‘Not wanting to be a burden’ could be a theme, because it says how participants make sense of supervision. For Braun and Clarke, themes are outputs of the analysis, not inputs to it.

Frequency does not make a theme either. Braun and Clarke advise against reporting how many participants said something: a pattern can matter without being common, and silence on a point is not disagreement. Their guidance is that two to six themes, subthemes included, is about right for an undergraduate project or a master’s dissertation.

Invented example: from codes to one theme

The study, participants and quotes below are invented. The research question is: How do international master’s students experience group work at a Danish university? Ten participants were interviewed and the transcripts coded in two rounds.

Invented example: five codes, a topic summary and a theme

Code	Semantic or latent	Extract
Waits to be asked before speaking	Semantic	‘In meetings I wait until someone asks what I think.’ (P2)
Watches home students to learn the norms	Semantic	‘I just watched what the Danish students did and copied it.’ (P1)
Agrees in the meeting, rewrites later	Semantic	‘I said yes in the meeting and then rewrote my part at night.’ (P7)
Informality experienced as a test	Latent	‘Everyone says it is relaxed, but you can still get it wrong.’ (P4)
Silence as self-protection	Latent	‘If I say nothing, at least I cannot say the wrong thing.’ (P9)

First attempt, phase three: ‘Experiences of group work’. This candidate theme gathered every extract about group work, including room bookings. In phase four it did not hold together: the extracts shared a topic but no central organising concept.

Developed theme, phase five: ‘Holding back until I know the rules’. Most participants experienced group work as governed by unwritten rules that nobody explained. Until they had worked those rules out, mostly by watching, they held back: they waited to be asked, agreed in meetings and made changes afterwards. Silence felt safer than breaking a rule.

Boundary and a contrasting case. The theme is about not yet knowing the rules, not about shyness. One participant who had done group work on a similar programme spoke up from the first meeting, which fits that reading.

The codes stay close to single extracts, while the theme name says what the pattern means. Keep the interview and timestamp beside each extract, so you can return to the context.

Quality: why inter-rater reliability does not fit

Inter-rater reliability assumes that a dataset has one correct coding for independent coders to reproduce. Reflexive TA rejects that assumption. As Braun and Clarke put it on their website, coding ‘can’t be right or wrong, but it can be weaker (superficial) or stronger (complex, nuanced, insightful)’. Agreement mainly shows that two people were trained to code alike.

So you do not need a second coder. In One size fits all? What counts as quality practice in (reflexive) thematic analysis? (2021), Braun and Clarke state that a research team is ‘not required or even desirable’ for quality in reflexive TA. Talking your codes through with your supervisor still helps, because it can reveal assumptions and readings you missed; report it as discussion, not agreement. In To saturate or not to saturate? (2021), Braun and Clarke argue that data saturation fits the logic of coding reliability TA, not reflexive TA, so justify your sample by your aims and the richness of your interviews instead.

To judge your analysis, use the 15-point checklist from the 2006 article, updated in the 2022 book. Common problems the authors describe include:

data that are paraphrased or summarised rather than analysed;
long extracts with little analytic narrative between them;
themes that overlap, or whose relationship is unclear;
claims that themes emerged from the data;
mixing approaches with different philosophies, such as reflexive TA with consensus coding.

Writing up reflexive thematic analysis in your dissertation

In the methodology chapter, name reflexive TA and cite the texts you followed. Then describe what you did rather than listing the six phases: how many rounds of coding, whether your coding leaned inductive or deductive and semantic or latent, and how your candidate themes changed. Say how your position may have shaped the analysis, and leave out inter-rater scores and claims of saturation.

In the findings chapter, organise the analysis by theme, not by participant or interview question, and open with a thematic map or a short table of themes. For each theme:

state its central organising concept in a sentence or two;
use extracts to show the pattern, and say what each one shows, either as an illustration or by analysing it closely;
show variation and contrasting cases, not only the clearest examples;
connect the theme to your research question and, where your programme expects it, to the literature.

Some dissertations combine findings and discussion in one chapter; others keep them apart, so follow your programme’s handbook. Label each quote with a pseudonym or participant number, mark cuts with […], and check every quote against the recording before you submit your dissertation or thesis.

Frequently asked questions

What is reflexive thematic analysis?: Reflexive thematic analysis is Braun and Clarke’s approach to thematic analysis, first set out in 2006 and named reflexive in 2019. You code qualitative data, generate themes as patterns of shared meaning held together by a central organising concept, and interpret them in relation to your research question. Your subjectivity is treated as a resource for the analysis, not as bias.
What are the six phases of thematic analysis?: In Braun and Clarke’s 2022 wording, the phases are familiarising yourself with the dataset; coding; generating initial themes; developing and reviewing themes; refining, defining and naming themes; and writing up. The 2006 article used older names for some of them, such as ‘searching for themes’ instead of ‘generating initial themes’ and ‘producing the report’ instead of ‘writing up’. The process is recursive, so you move back and forth between phases.
How is reflexive thematic analysis different from codebook thematic analysis?: Codebook approaches, such as template and framework analysis, use a structured coding frame and usually develop themes early, often as summaries of what participants said about each topic. Reflexive TA uses no codebook: codes evolve throughout coding, and themes are developed late, as the outcome of coding and interpretation. Both differ from coding reliability approaches, which measure agreement between independent coders.
Do I need a second coder for reflexive thematic analysis?: No. Braun and Clarke argue that coding in reflexive TA is interpretation, so there is no correct coding for a second coder to confirm, and inter-rater reliability scores do not fit the approach. Discussing your coding with your supervisor is still useful, as long as the aim is to reflect on your assumptions and readings rather than to reach agreement.