Skip to main content

Step-by-step guide

How to Do Thematic Analysis: A Step-by-Step Guide

This is a practical, step-by-step guide to doing thematic analysis on qualitative data — most often interview transcripts. It follows Braun and Clarke's six-phase framework and adds the documentation each phase needs to satisfy an examiner or reviewer.

Work through the phases recursively: you will move back and forth between coding and theme development as your understanding deepens. That is expected, not a sign of doing it wrong.

The six phases, step by step

  1. 1Familiarise yourself with the data — transcribe (or check transcripts), then read and re-read the full dataset. Note early impressions and recurring ideas before you start coding — this is your first analytic pass, not a formality.
  2. 2Generate initial codes — go through the data systematically and tag every segment that is interesting or relevant to your question with a short code. Code inclusively (you can discard later), code for as many patterns as you notice, and keep a record of what each code means.
  3. 3Search for themes — cluster related codes into candidate themes — broader patterns of meaning. A theme is not just a topic; it captures something meaningful about the data in relation to your research question. Collate all coded data relevant to each candidate theme.
  4. 4Review the themes — check each theme against its coded extracts (does it hold together?) and against the whole dataset (does it reflect the data overall?). Refine, merge, split, or discard. A theme that is really two ideas should be separated; thin themes may be collapsed.
  5. 5Define and name themes — write a clear definition for each theme — what it is and is not — and give it a concise, informative name. Identify the story each theme tells and how the themes fit together into an overall narrative.
  6. 6Produce the report — select vivid, representative extracts (spanning multiple participants, not one voice), weave them into an analytic narrative that answers your research question, and relate the analysis back to the literature.

How to code interview data well

Coding is where most of the analytic work happens. Two practical habits make a difference: code across the entire dataset before building themes (so themes are grounded in all the data, not the first few transcripts), and keep your codes anchored to actual extracts so you can always trace a theme back to its evidence.

Watch for over-reliance on a single participant — examiners flag themes that rest on one vivid voice. Aim for patterns supported across multiple participants.

Documenting rigour as you go

The difference between a pass and a revision is often documentation, not analysis. As you work, keep: a codebook showing how codes developed and what they mean; an audit trail of decisions (codes created, merged, renamed); a reflexivity record of how your position shaped interpretation; and — if you used AI — a disclosure of where it assisted and where you decided.

QualIntel OS maintains the codebook and audit trail automatically as you accept or reject AI-surfaced candidate evidence, and generates the AI disclosure statement from that log — so the rigour you did is documented without extra admin.

Frequently asked questions

How do you do thematic analysis step by step?

Follow six phases: (1) familiarise yourself with the data by reading it thoroughly, (2) generate initial codes across the whole dataset, (3) search for themes by clustering related codes, (4) review themes against the data, (5) define and name each theme, and (6) write the report using representative extracts. The phases are recursive — expect to move back and forth.

How long does thematic analysis take?

It varies with dataset size and depth, but coding and theme development for a typical postgraduate interview study (15–25 transcripts) often takes several weeks of focused work. AI-assisted evidence retrieval can speed the coding phase considerably, provided you still confirm every decision — which is what keeps the analysis yours and defensible.

How many themes should a thematic analysis have?

There is no fixed number, but most studies report somewhere between two and six main themes, often with subthemes. Too many themes usually means you are reporting topics rather than developing analytic patterns; too few may mean themes are too broad. Let your research question and data guide it, not a target count.

How do I show rigour in thematic analysis?

Document how themes were developed: keep a codebook, an audit trail of coding decisions, and a reflexivity statement, and report against a standard (Braun & Clarke's guidance, or COREQ/SRQR for the wider study). If AI assisted, disclose its role. QualIntel OS produces the codebook, audit trail, and disclosure automatically.

Related guides

Run your thematic analysis in QualIntel OS

AI surfaces candidate evidence, you confirm every decision, and the audit trail writes itself. One project free, no card.

Start for free