Thematic Analysis in Qualitative Research: The 6-Step Framework

Most of the work in a Braun and Clarke thematic analysis is interpretive. AI cannot shortcut that part. Boyatzis formalized thematic coding and Braun and Clarke later refined it into the dominant analytic framework in academic and commercial qualitative research. AI automates several mechanical steps, but only rigorous human oversight turns that automation into usable insight.

Key Takeaways

• Braun and Clarke's six-step thematic analysis loop (familiarize, code, cluster, review, define, report) is the field's standard framework across academic and commercial research
• A twenty-interview study consumes roughly 80 analyst hours; a hundred-interview study at the same depth would consume 400, which is why large-sample qual was historically avoided
• AI can run initial coding across a full transcript corpus in minutes and query passages like a database, compressing the mechanical layer of analysis dramatically
• A plain LLM asked to "summarize the themes" produces fluent output that is wrong in specific and identifiable ways. Structured codebooks and human review are essential
• The researcher's job shifts toward codebook design, cluster review and narrative construction

The Six-Step Loop and the Reflexive Revision

Braun and Clarke's framework runs in six steps:

1. Familiarization. Read every transcript several times to get to know the data.
2. Initial coding. Tag passages with short descriptive labels.
3. Theme search. Cluster those codes into higher-order patterns.
4. Theme review. Check each theme against the full corpus to confirm it holds.
5. Defining and naming. Pin down what each theme covers and give it a precise name.
6. Writing up. Build the analytic narrative with supporting verbatims.

Most commercial researchers recognize this loop even when they haven't named it. Where they diverge is in how rigorously they execute it. Perfect fieldwork analyzed poorly produces a report that is fluent, well-quoted and useless. For a deeper look at how interview design feeds into the analysis phase, see Depth Interview Design: The Three Layers Every Researcher Must Know.

In later work Braun and Clarke recast their 2006 framework as reflexive thematic analysis. The change matters for commercial practice. Many readers had treated the six steps as a procedural recipe. The reflexive version treats the process as interpretive: a researcher constructs each theme by making active choices at every step.

Braun and Clarke have also argued against mechanical applications of their framework, including inter-rater reliability scores used as a quality check. In their view coding is interpretation, so a reliability coefficient misrepresents what good analysis looks like.

For applied research, reflexive thematic analysis demands a different kind of accountability. The researcher documents interpretive decisions; that documentation is what makes a theme defensible when a CMO pushes back. Enumerate treats every generated theme as a first pass. The researcher reviews it and can rename or remove it before it applies to the data.

The Throughput Ceiling That Kept Qual Small

The economic constraint on qualitative thematic analysis has always been analyst time. A twenty-interview study coded to the depth the framework demands can take roughly 80 hours. That works out to about four hours per interview.

Scale that to a hundred interviews and the hours reach 400, about ten weeks of one analyst's working time. This is why large-sample qualitative work was rarely run, even when a study had several segments that each needed to reach saturation.

How AI Fits Into the Six-Step Loop

A carefully prompted AI system can run initial thematic coding across a full corpus in minutes. It can also auto-cluster codes by semantic similarity and retrieve every passage where a respondent raised a specific concern, much like querying a database. These are real capabilities that compress the mechanical layer of the six-step loop dramatically.

The failure mode is equally real. A plain LLM handed transcripts and asked to summarize themes produces fluent output that is wrong in specific ways. It weights themes by how often they appear, smooths over contradictions and cannot tell when the research question was the wrong one. Automated Coding for Qualitative Data covers the implementation detail.

Codebook design, cluster review and verbatim selection remain irreducibly human. The researcher defines the categories the machine searches within. They also catch when two similar codes capture distinct phenomena and decide which quote earns its place in the final narrative. Interpretive authority stays with the analyst.

Want to see AI-assisted thematic analysis running on a live study? Book a demo with Enumerate.