A single participant can mention the same problem several times. If the claim is about participants, deduplicate coded segments by participant ID before counting; if it is about mentions, say so and keep that separate measure. Preserve the eligible interview denominator, missing records, and recruitment scope. Qualitative interviews can show mechanisms and variation, but a small purposeful sample does not produce a population prevalence estimate. The prompt may compute descriptive counts from an explicit ledger, while a deterministic check reconciles those counts to distinct participant IDs.
Make this comfortable
Interview prompts: count people, not repeated excerpts
Operational case
The Parcel Window ledger has 47 coded excerpts from 12 participants. Nine excerpts about window clarity come from five distinct participants; four of those excerpts are P-07 revisiting one event. Reporting '9 of 12 customers' would count text spans as people and overstate the finding. The study can say five of the 12 interviewed participants raised unclear windows, with the recruitment context nearby. Two participants mention a language barrier, but their remarks remain a separate finding rather than being folded into window clarity.
Eligible interviews: 12
Coded excerpts: 47
WINDOW_CLARITY excerpts: 9
Distinct WINDOW_CLARITY participants: 5
LANGUAGE_BARRIER participants: 2
Population prevalence: not estimatedPerformance and review cost
Deduplicating M coded excerpts with a participant-ID set is O(M) expected time and O(P) memory for P distinct participants. Manual verification remains necessary for misassigned speakers and duplicated interview records. Descriptive counts are cheap; false precision is costly because it can make a narrow exploratory study look like a survey. Preserve both segment and participant counts so a reviewer can reproduce the statement.
Common Mistakes
- Do not divide quote counts by participant counts.
- Do not call 5 of 12 a population rate.
- Do not combine separate codes merely to make a larger number.
Connected lessons
- Production prompt engineering
- Prompt Engineering
- Aggregation prompts: pin the denominator and recompute the rate
- Numeric prompts: let code calculate and the model explain
- Interview prompts: limit transcript use to the study purpose
- Interview prompts: preserve speaker and segment boundaries
- Interview prompts: write a codebook with inclusion rules
- Interview prompts: keep coder disagreement visible
- Interview prompts: build a theme evidence ledger
- Interview prompts: turn bounded findings into testable decisions
- Project: analyze Parcel Window pickup interviews
- Qualitative interview prompt decisions
prompt engineering
qualitative research