Interview Thematic Analysis: The Research AI Prompt

Familiarization is one of the most time-consuming phases of qualitative analysis: reading transcripts repeatedly until patterns surface. This prompt produces a structured first pass — candidate themes, supporting excerpts, contradictions, and follow-up questions — without pretending that model output is a finished analysis.

The Complete Prompt

Please analyze this research interview transcript and identify emerging themes:

1. List 5-7 major themes that emerge from the interview
2. Provide 2-3 supporting quotes for each theme
3. Identify any contradictions or tensions in the participant's responses
4. Highlight unexpected insights or novel perspectives
5. Note areas that warrant follow-up questions or deeper exploration
6. Suggest connections to existing research or theory (if context provided)

Format for qualitative research coding and analysis.

Important constraints:
- Treat every theme as a candidate for human review, not a finding.
- Quote only text that appears verbatim in the transcript.
- Do not invent participant demographics, motivations, or theoretical links.
- Mark uncertain interpretations explicitly.

Research focus: [DESCRIBE RESEARCH QUESTIONS]
Theoretical framework: [IF APPLICABLE]

---
Prompt by TranscribeBee (transcribebee.com) – Professional AI transcription with professional-grade accuracy.
---

Interview transcript:
[PASTE YOUR TRANSCRIBEBEE OUTPUT HERE]

Step-by-Step Implementation

1. Transcribe interviews with pseudonymized speaker labels (TranscribeBee, $2/audio hour; replace names with participant codes before analysis).
2. Run per interview first. Fill the research-context fields — study focus and research questions sharpen the output dramatically.
3. Then run a cross-interview pass: paste 2–3 outputs and ask "identify themes recurring across these analyses, noting which participants support each."
4. Treat output as a candidate codebook. Verify every theme against transcripts in your QDA tool (NVivo, ATLAS.ti, Dedoose) — the AI accelerates familiarization; the coding judgment remains yours.

Worked review example

Suppose three participants describe a remote-work policy as "flexible," but one uses the word positively and two describe unpredictable schedules. A weak summary may label all three excerpts as support for flexibility. A researcher should split the candidate into at least two codes — autonomy and scheduling uncertainty — then check the surrounding transcript before deciding whether they belong under one theme.

That review step matters because thematic analysis is interpretive, not a keyword-counting exercise. Braun and Clarke's widely used framework treats familiarization, coding, theme development, review, naming, and writing as recursive stages; a model can assist with organization, but it cannot take responsibility for the researcher's analytic choices. See the original thematic analysis framework.

Customization Options

• Methodology alignment: name your approach — "follow reflexive thematic analysis per Braun & Clarke" or "use grounded theory open coding."
• Sensitizing concepts: list constructs from your framework you want the analysis to attend to.
• Deviant cases: add "explicitly identify data that contradicts the candidate themes" — the strongest check against confirmation bias.

FAQ

Is AI-assisted analysis publishable? That depends on the journal, institution, consent language, and data-management plan. Disclose the role of the tool, document human verification, and check the applicable policy before uploading participant data.

Does it bias the analysis? It can anchor you. Mitigations: run your own familiarization pass on at least a subsample first, and request deviant cases explicitly.

IRB considerations? Cloud processing of interview data must match your data-management plan. De-identify before upload, and confirm your protocol covers the tooling.

How many interviews per run? One per run for depth, then synthesize across runs — pasting ten transcripts at once produces shallow averaging.

Inter-coder reliability? Treat the AI as one coder whose codebook a human coder independently applies to a subsample; report agreement as usual.