Thematic Analysis

Thematic Analysis turns transcripts into themes, then synthesizes across participants. It follows a four-phase inductive process per interview, then a cross-case pass over the whole study.

September 26, 2026

Where it is

Thematic in the sidebar (/thematic-analysis/<study id>). Three tabs:

  1. Setup — choose the model, review credits, watch progress
  2. Analyze — run the four phases on individual interviews
  3. Synthesize — cross-case analysis and exports

Only interviews that are completed can be analyzed.

Setup

Model Selection decides quality and cost per analysis:

Tier Model Cost Availability
Open Weights DeepSeek V4 Flash 1 credit All plans
Standard Kimi K2.7 3 credits Solo and up
Premium Claude Sonnet 5 10 credits Pro and up

The credit figure in that table is the model's rate per five questions of the study. An analysis is then charged at 2× that rate, so a 5-question study analysed on Open Weights costs 2 credits per interview, and a 12-question study costs 6 (three blocks of five, doubled).

The Credit Estimation panel does this arithmetic for you: questions per interview, credits per analysis, pending interviews, estimated total and your balance, all before you start. If you are out of credits the page says so and links to Billing.

The counters at the top read: total interviews with completed responses, how many have all four phases done, and how many are awaiting analysis.

The four phases

Each interview goes through:

  1. Initial Engagement — multiple readings (3 runs) to capture diverse first impressions and flag areas of confusion
  2. Initial Noting — annotate each message with descriptive [D], linguistic [L] and conceptual [C] codes, noting contradictions and ambiguities
  3. Emergent Themes — turn annotations into a theme inventory with titles, descriptions, keywords, supporting quotes and interpretative notes
  4. Connections — identify patterns, relationships, superordinate themes and contradictions, producing a thematic structure

Phases run in order: phase 3 needs phase 2, phase 4 needs phase 3.

Start Analyzing Interviews on the Setup tab does not start anything by itself — it takes you to the Analyze tab. There you pick an interview from Select Interview and click Analyze Interview, which runs all four phases in sequence; you can also run each phase on its own and read its output before paying for the next. Phase 2 can take several minutes on a long transcript.

If a phase fails, the step turns red and the message says which phase and why. Credits reserved for a failed phase are released, so a failure costs you nothing.

Each phase's output is shown in a results pane with fullscreen and word-wrap toggles, including the full annotated transcript.

Export Interview Analysis gives you PDF or Markdown once all four phases are complete.

Cross-case synthesis

The Synthesize tab looks across analyzed interviews for shared experiences and unique variations. It needs at least 3 completed interview analyses.

Exports:

  • Export Study Results — the synthesized findings
  • Export for AI Agents — a requirements-style Markdown document derived from the cross-case analysis, meant to be read by a coding agent
  • Research Presentation — a PowerPoint summarizing the findings, generated from the cross-case analysis

Practical advice

  1. Analyze a couple of interviews first and read the phase outputs before spending credits on the whole set — it is the cheapest way to find out the transcripts are too thin.
  2. Model choice matters more here than in interviewing. Phase 3 and 4 are interpretive work; the open-weights tier is fine for a first look, less so for findings you will present.
  3. Three interviews is a floor, not a target. Cross-case output gets meaningfully better with more.
  4. Read the quotes. The value of the export is the supporting quotes, which tell you whether a theme is real or an artifact.

Known limits

  • Only completed interviews are eligible. A study whose interviews all ended early has nothing to analyze.
  • There are no MCP tools for thematic analysis. Earlier documentation described run_thematic_analysis, get_analysis_status and get_agent_requirements; those tools do not exist in the current server. An agent can reach the raw material with bulk_transcripts and search_transcripts and do its own analysis, but the four-phase pipeline and its exports are dashboard-only.

Last updated: September 2026