You have run the interviews. You have forty pages of transcripts, a head full of half remembered quotes, and a stakeholder asking what you found. Thematic analysis is how you get from that pile to a clear, defensible set of themes, without quietly cherry picking the quotes that confirm what you already believed. This is a step by step walkthrough you can actually follow.
What thematic analysis is
Thematic analysis is a method for finding patterns of meaning across qualitative data: interviews, open survey responses, support tickets, diary entries. You read the data closely, label the meaningful bits, and group those labels into themes that tell you something true about the whole set.
It is the most widely used approach to qualitative analysis for a reason: it is flexible, it does not demand a particular theoretical stance, and it scales from a handful of interviews to a large study. It sits firmly on the qualitative side of the qualitative and quantitative divide: depth over counts. The version most people reference is the six phase approach described by Braun and Clarke, and that is the spine of this guide.
Inductive or deductive: decide your starting stance
Before you code a single line, decide where your themes will come from.
- Inductive (bottom up): themes emerge from the data itself. You come in with as few preconceptions as you can manage and let the patterns surface. Best for exploratory work.
- Deductive (top down): you start with a framework or set of questions and code against it. Best when you have specific questions to answer or are comparing against a model.
Most real analysis is a blend, leaning one way. Naming your stance up front keeps you honest about whether you are discovering themes or confirming them.
The six phases, step by step
1. Familiarise yourself with the data
Read everything, properly, before you code anything. If you have recordings, consider doing or checking the transcription yourself, because the act of transcribing is where a lot of noticing happens. Jot down early impressions, but hold them loosely. The goal of this phase is immersion, not conclusions.
2. Generate initial codes
Now go through the data systematically and label anything meaningful. A code is a short tag that captures what a chunk of data is about, for example "fear of getting it wrong" or "workaround for slow export."
Practical habits:
- Code at the level of an idea, a phrase or a few sentences, not a whole paragraph
- Keep codes close to the participant's own meaning, especially when working inductively
- It is fine for one piece of data to carry several codes
- Code generously in the first pass; you can merge later
By the end you will have a long, messy list of codes. That is exactly what you want.
3. Search for themes
Step back and look for how the codes cluster. A theme is bigger than a code: it captures a pattern that says something significant about your research question. Group related codes together and give each cluster a working name.
This is where physical or digital affinity mapping earns its keep. Spread the codes out, move them around, and let groupings form. Some codes will become themes, some will become sub themes, and some will turn out to be noise.
4. Review themes
Now pressure test what you have, at two levels:
- Against the coded data: does every theme actually hold together, with enough evidence behind it? Split themes that are doing too much; merge ones that overlap; drop ones that are thin.
- Against the whole data set: re read everything with your candidate themes in mind. Do they capture the data accurately, or did you miss something that does not fit?
A good theme is coherent (the bits inside belong together) and distinct (it does not blur into the theme next door).
5. Define and name themes
For each theme, write a couple of sentences capturing what it is, what it includes, and why it matters. If you cannot describe a theme cleanly in a sentence or two, it is probably two themes, or not a theme at all.
Give themes names that carry meaning. "Onboarding" is a topic. "Onboarding feels like a test they are afraid to fail" is a theme: it says something.
6. Write up
Tell the story your themes reveal, and back each one with evidence: a clear claim, then a representative quote or two. Show the reader the data, do not just assert the pattern. Where it helps, note how common a theme was across participants, while being careful not to dress qualitative depth up as false precision.
How to stay honest
Thematic analysis can be rigorous or it can be elaborate confirmation bias. A few guardrails keep it on the right side:
- Code before you conclude. If you already "know" the findings, you will code your way to them.
- Count participants, not quotes. Five vivid quotes from one talkative person is not a theme. Note how many participants a pattern actually spans.
- Keep an audit trail. Anyone should be able to follow a theme back down to the codes and the raw data behind it.
- Look for the disconfirming case. Actively hunt for data that contradicts your theme. If you cannot find any, you may not have looked.
Where AI fits, and where it does not
AI tools can genuinely help with the mechanical parts: transcribing, suggesting an initial pass of codes, clustering similar snippets. Used well, they save hours.
What they cannot do is own the meaning. The judgement about what a theme is, what matters, and what a participant really meant in context still has to be yours. Treat AI as a fast, slightly naive research assistant whose work you always check, not as the analyst.
The skill underneath the steps
The six phases are a scaffold, not a recipe you can follow without thinking. The real skill in thematic analysis is reading closely, holding your assumptions loosely, and being willing to let the data tell you something you did not expect. Do that, follow the phases, and you will produce findings you can stand behind when someone asks "how do you know?"
People of Research is a community and newsletter for UX researchers, behavioural scientists, and anyone who turns messy human data into clear decisions. Join us and get the good stuff in your inbox.