Coding requires a stable unit of evidence. Split transcripts into segments that retain participant ID, speaker, interview question, time span, and a segment identifier. Keep enough surrounding turns to resolve pronouns and corrections, but do not join different participants into one fabricated quote. An automatic transcript may misassign a speaker or omit a negation; mark such spans uncertain and compare them with the recording when available. The model can propose segment boundaries, while the researcher resolves uncertain attribution. Each downstream code and finding should point back to the approved segment version.
Make this comfortable
Interview prompts: preserve speaker and segment boundaries
Operational case
At 12:14, P-03 says a pickup window was 'fine' and immediately corrects that statement after describing a missed collection. A one-sentence clip would reverse the meaning. The analyst keeps the correction in segment P03-S18, records the time range, and marks a noisy word for audio review. P-08's later complaint about the same screen stays in its own segment; merging those two voices would falsely suggest one person described both events.
P03-S18 | speaker P-03 | 12:14-12:42
Initial: window was fine
Correction: collection missed after the window changed
Audio uncertainty: one word pending review
P08-S09: separate participant, separate segmentPerformance and review cost
Segmenting T transcript turns is O(T) before any comparison between segments. Wider context improves interpretation but increases token cost, so attach nearby turns only where they change meaning. Versioning a corrected transcript adds storage but prevents a code from pointing at text that no longer exists. A model output without participant and segment keys is cheap to produce and expensive to audit.
Common Mistakes
- Do not code the first sentence while dropping its immediate correction.
- Do not merge similar remarks from different participants into one quote.
- Do not silently repair uncertain speech without checking the recording.
Connected lessons
- Prompt engineering applications
- Prompt Engineering
- Audio prompts: preserve speakers, timestamps, and uncertain words
- Caption and transcript prompts: keep timing, speakers, and sound evidence
- Interview prompts: limit transcript use to the study purpose
- Interview prompts: write a codebook with inclusion rules
- Interview prompts: keep coder disagreement visible
- Interview prompts: build a theme evidence ledger
- Interview prompts: count people, not repeated excerpts
- Interview prompts: turn bounded findings into testable decisions
- Project: analyze Parcel Window pickup interviews
- Qualitative interview prompt decisions
prompt engineering
qualitative research