Marcus Andrey Study
Ler em português Library

UX Research · Step 6 · 8 min read

Thematic analysis

Thematic analysis is the most widely used way to find patterns of meaning in qualitative data, and the most often misused. What codes and themes are, the six phases, why themes don't 'emerge', the difference between a theme and a topic summary, and how to do it well in a design team.

A theme isn't something you find in the data. It's something you make from it, carefully, and can defend. That sounds like a philosophy-class point, but it changes how you work. If themes were waiting in the transcripts, anyone would find the same ones and you'd only need to look. Because you make them, you need a method that keeps you honest: code everything, check every theme against the data, and say what you brought to the reading.

What you'll be able to do after reading:

  • Explain what codes and themes are, and how they differ
  • Code an interview excerpt with descriptive and interpretive codes
  • Work through the six phases of thematic analysis
  • Tell a real theme from a topic summary

1. What it is

Virginia Braun and Victoria Clarke's 2006 paper started by calling thematic analysis "a poorly demarcated, rarely acknowledged, yet widely used qualitative analytic method", and set out to give it clear guidelines. They argued it's accessible and theoretically flexible, useful well beyond psychology (Braun and Clarke, Qualitative Research in Psychology, 2006). Semantic Scholar counts more than 90,000 citations of it (Semantic Scholar), and its ideas sit behind a lot of UX synthesis, even when nobody names them.

NN/g's definition is practical: a systematic method of breaking down and organising rich qualitative data by tagging observations and quotes with codes, to help discover significant themes (NN/g, Thematic Analysis, 2022). Two words carry it:

  • A code is a word or phrase that labels a segment of text: shorthand for what it's about, like a hashtag on a post.
  • A theme is a pattern of shared meaning across the data, held together by one central idea. In Braun and Clarke's words today, themes are "pattern of shared meaning underpinned by a central concept or idea" (thematicanalysis.net).

2. Coding

P4, interview 4, 18:40

Honestly, I only use the app for the big stuff, like rent.1 For the small things I just let it go, it's like, five euros, who cares.2 But then it adds up and suddenly it's sixty3 and I have to say something, and that's awkward.4

I turned the reminders off5 because it felt like the app was me, nagging Ana.6 If it just came from the app, on the first of the month, like a bill, I wouldn't mind.7

  1. 1Big amounts onlyDescriptive
  2. 2Lets small debts goDescriptive
  3. 3Small debts build upDescriptive
  4. 4Asking for money feels like conflictInterpretive: what it means to him
  5. 5Turned reminders offDescriptive
  6. 6The reminder speaks in my voiceInterpretive: why he turned them off
  7. 7Wants a scheduled, bill-like reminderDescriptive
Ninety seconds of one interview, coded. Highlighted segments get descriptive codes, what the text is about; underlined ones get interpretive codes, what it seems to mean. Codes 2, 3, 4 and 6 later feed the theme "asking for money feels rude, so people wait".Diagram: Marcus Andrey, CC BY 4.0. A made-up interview. Descriptive and interpretive codes as described in NN/g, Thematic Analysis (Rosala, 2022).

Coding is going through the data a segment at a time and asking "what is this about?" NN/g separates two kinds of code (NN/g, 2022):

  • Descriptive codes say what the text is about, close to the participant's words: "turned reminders off".
  • Interpretive codes add your analytical reading: "the reminder speaks in my voice". They usually come more easily later, once you've read everything.

Braun and Clarke describe the same spectrum as semantic (the explicit content) to latent (the ideas and assumptions underneath) (thematicanalysis.net). Both are useful. A study with only descriptive codes tends to end in a list of what people said; a study with only interpretive codes tends to drift away from the data.

Keep a code list as you go, with a short description and an example of each code. NN/g notes it saves you from creating three codes for the same thing, and it's essential when more than one person codes.


3. The six phases

  1. Familiarise yourself

    Read and re-read the whole data set, and note first observations.

  2. Code

    Give short labels to everything relevant to the research question; two rounds or more.

  3. Generate initial themes

    Build candidate patterns of shared meaning from the codes.

  4. Develop and review

    Check each theme against its coded extracts and the whole data set; split, merge or drop.

  5. Refine, define and name

    Work out each theme's scope, its story and an informative name.

  6. Write up

    Weave the analysis and the extracts into one story.

Sequential, but recursive: you go back and forth between phases.

The six phases of reflexive thematic analysis, with the names Braun and Clarke use today. The names have changed since their 2006 paper; the idea of six phases hasn't.Diagram: Marcus Andrey, CC BY 4.0, after Braun and Clarke, Doing reflexive TA (University of Auckland).

Braun and Clarke's six phases, in the names they use now (Doing reflexive TA):

  1. Familiarise yourself with the data set. Read and re-read everything, even the sessions you ran yourself. NN/g suggests working from transcripts, not memory.
  2. Code. Code the whole data set, in two or more rounds, then collate the codes and their extracts.
  3. Generate initial themes. Look across the codes for broader patterns of meaning, and gather the data for each candidate.
  4. Develop and review themes. Check each candidate against its extracts, then against the whole data set. Split, merge or drop.
  5. Refine, define and name themes. Write what each theme is about and what it isn't, and give it a name that says something.
  6. Write up. Weave the story of the themes together with extracts.

They stress that the phases are sequential but recursive: you move back and forth. NN/g adds a step teams skip: take a break and come back to your themes with fresh eyes before you finalise them.

The Six phases of thematic analysis card has the phases as steps, a worked example and a template you can copy.


4. Themes are made, not found

In their 2019 reflection on the 2006 paper, Braun and Clarke clarified how their thinking had changed, and said they now prefer to call their approach reflexive TA (Braun and Clarke, 2019). Its core idea: themes are generated by the researcher, through familiarisation, coding and theme development, and coding "inevitably and inescapably bears the mark of the researcher(s)" (thematicanalysis.net, FAQs).

Three consequences for a design team:

  • Say "we developed three themes", not "three themes emerged". It's more honest, and it reminds everyone that the team's assumptions shaped the reading.
  • Write down what you brought in. Before you code, list what you expect to find and what you hope to find. When a theme matches your hopes exactly, look at it twice. How people think covers confirmation bias.
  • Agreement scores aren't the goal. Some approaches measure how often two coders agree. Braun and Clarke argue that in reflexive TA, with no single "accurate" coding, the logic of inter-coder reliability disappears; their 2021 paper criticises treating it as a universal quality requirement (Braun and Clarke, 2021). Two people coding together is still valuable, for discussion, not for a score.

They also describe thematic analysis as a family of approaches, not one method: coding reliability, codebook and reflexive approaches differ in philosophy and procedure (thematicanalysis.net). In product teams, a light codebook approach is common when several researchers share a repository. Whatever you use, say which.


5. A theme or a topic summary?

The most frequent problem Braun and Clarke see in published work is the topic summary presented as a theme: a summary of everything participants said about one topic or one interview question, with nothing that ties it together (thematicanalysis.net, FAQs). Design reports are full of them.

Topic summary Theme
"Reminders": some people like them, some turned them off, some want them weekly Asking for money feels rude, so people wait, and a reminder in their own voice makes it worse
"Onboarding": feedback on the sign-up screens People skip setup because they only open the app when a bill is already due

A quick test: if the theme's name could be an interview question, or a section of your discussion guide, it's probably a topic. A theme has a central idea, and you could explain it to someone in one breath.


6. Doing it in a design team

  • Pick the tool to fit the data. NN/g describes three: qualitative-analysis software, journaling with memos, and affinity diagramming with cut-up transcripts (NN/g, 2022). For six interviews, a spreadsheet with one row per coded segment is enough.
  • Involve the team in familiarisation. NN/g suggests a workshop where everyone reads and highlights transcripts, passing them on until each person has read all of them. It fixes the problem of colleagues who watched one session and remember only that one.
  • Keep the codes traceable. Each coded segment keeps its participant and timestamp, so every theme in the report can point back to the moment it came from.
  • Use AI as an assistant, not an analyst. It can draft codes for you to check; it can't take responsibility for the reading. AI-assisted synthesis, step by step shows a careful workflow.

7. Common mistakes

  • Skimming the transcripts and coding only the memorable bits.
  • Codes so broad they're topics ("Reminders").
  • Themes named like interview questions.
  • "Themes emerged", as if nobody did anything.
  • Counting how many times a word appears and calling it analysis.
  • Ignoring the data that contradicts a theme. NN/g lists ignoring viewpoints that don't fit the researcher's beliefs among the common problems.
  • Stopping at codes, so the report is a list of what people said.

8. Vocabulary

Term Meaning
Code A short label for a segment of data
Coding Labelling segments of the data with codes
Descriptive or semantic code A label for what the text says
Interpretive or latent code A label for what the text seems to mean
Code list (codebook) The codes, each with a description and examples
Theme A pattern of shared meaning, held together by a central idea
Topic summary Everything said about one topic, without a central idea
Reflexive TA Braun and Clarke's approach, where the researcher's role is acknowledged
Reflexivity Noticing and saying how you shaped the analysis

9. Try it (no tools needed)

  1. Take the coded excerpt in the figure and cover the codes.
  2. Code it yourself: three descriptive codes, one interpretive.
  3. Compare with the figure. Where you differ, which reading can you defend better from the words?

Reflection questions: What did you expect P4 to say before you read it? Did that expectation change your codes?


10. If you already practise

  • Write a short reflexivity note at the start of every study: what you expect, what you'd like to be true, and what role you play in the product.
  • Keep a code list in the study folder, with descriptions and examples, and reuse it in the next study of the same area.
  • Review themes in pairs before the report, asking of each one: what's the central idea? What data argues against it?
  • Read Braun and Clarke's twenty critical questions in their 2021 paper; they're a good checklist for your own reports.

Checklist

Learn more

Books, courses and videos to go deeper

Book links go to Amazon. As an Amazon Associate I earn from qualifying purchases, at no extra cost to you.

From notes to themes, affinity mapping · Personas and journeys from data · Notes, recordings and transcripts · AI-assisted synthesis, step by step · Synthesise a study

Saved only on this device.

↑ ↓ to move · Enter to open · Esc to close