Audit Qualitative Interview Transcripts for Researcher Bias

Get a turn-by-turn bias audit of your interview transcript, a list of contaminated participant quotes to exclude from coding, and neutral rewrites for every leading question.

You getBias Summary, turn-by-Turn Audit Table and excluded Quotes

Prompt

Prompt
Act as a senior qualitative research methodologist specializing in interview reliability.

I will paste an interview transcript right after this prompt. Audit every interviewer turn for researcher bias before I begin thematic coding.

Follow these rules:
1. Tag every interviewer turn as Clean, Leading, Double-Barreled, Loaded, Confirmation-Seeking, or Premature-Summary.
2. Flag any participant quote immediately following a biased turn as Contaminated if it echoes the interviewer's framing, and state why it should be excluded from coding.
3. Rewrite every biased turn into one open-ended neutral probe that keeps the research intent without suggesting an answer.
4. Quote the transcript verbatim with turn numbers or timestamps. Never invent lines or assume unstated context.

Output format:
1. Bias Summary: total interviewer turns, counts by tag, and contamination rate as a percentage.
2. Turn-by-Turn Audit Table: columns for Turn, Bias Tag, Original Question, and Neutral Rewrite.
3. Excluded Quotes: bulleted list of contaminated participant quotes with turn number and exclusion rationale.
4. Interviewer Habits to Fix: 2-3 bullets on recurring patterns for the next session.

End with one line noting missing context, such as the core research question or tone of voice, that could affect your audit.

Example output

At a glanceInterviewer turns by bias type4 of 14 turns biased; 3 participant quotes flagged for exclusion
  • Clean: 10
  • Leading: 2
  • Loaded: 1
  • Confirmation-Seeking: 1

Example for: a 25-minute UX research interview with a hospital nurse about a new shift-handover app.

1. Bias Summary
  • Total Interviewer Turns: 14
  • Clean: 10
  • Leading: 2 (Turns 5, 11)
  • Loaded: 1 (Turn 8)
  • Confirmation-Seeking: 1 (Turn 13)
  • Contamination Rate: 28.6% (4 of 14 turns biased; 3 participant responses flagged for exclusion)
2. Turn-by-Turn Audit Table
TurnBias TagOriginal QuestionNeutral Rewrite
Turn 5Leading"Did the cluttered alert screen slow down your morning handover?""How did the alert screen affect your morning handover, if at all?"
Turn 8Loaded"How frustrating is it when night-shift notes fail to sync?""What happens when night-shift notes do not sync, and how do you handle it?"
Turn 11Leading"Would a one-tap summary button make your job much easier?""What changes, if any, would improve how you review handover summaries?"
Turn 13Confirmation-Seeking"So it sounds like you mostly ignore the medication checklist, right?""How do you typically use the medication checklist during handover?"
3. Excluded Quotes
  • Turn 6 (Participant): "Yeah, it definitely slowed me down because it felt cluttered." — Exclude: participant directly adopted the interviewer's word "cluttered" and confirmed a suggested cause.
  • Turn 9 (Participant): "It is super frustrating, honestly." — Exclude: emotional framing was primed by the word "frustrating" in Turn 8.
  • Turn 12 (Participant): "Yes, a one-tap button would make it way easier." — Exclude: participant agreed with a hypothetical solution pitched by the interviewer.
4. Interviewer Habits to Fix
  • Premature solution pitching: Avoid testing specific UI ideas (like a one-tap button) until the participant has described their workflow pain points unprompted.
  • Emotional labeling: Replace adjectives like "cluttered" and "frustrating" with neutral prompts that let the participant choose their own descriptors.

Note on missing context: Audio tone, pauses, and the overarching study research question were not provided; sarcastic or hesitant delivery could change whether Turn 14 is also contaminated.

Written by AI from this prompt. Your results will vary.

Tips and common edits

Why this works

  • Forces a fixed taxonomy of six bias tags so the model audits every turn consistently instead of giving vague feedback.
  • Links biased questions directly to contaminated participant quotes, preventing tainted data from entering thematic coding.
  • Pairs every flagged turn with a neutral rewrite, turning the audit into an immediate training guide for the next interview.

Paste an anonymized interview transcript right after this prompt in Claude, Gemini, or ChatGPT before thematic coding. Include your core research question if you want tighter relevance checks.

Common edits

  • Paste your core research question above the transcript if you also want the model to flag off-topic tangents.
  • Add a rule to score each biased turn by severity (Low, Medium, High) when auditing long transcripts over 50 turns.

Make it repeatable

Keep using this prompt without pasting it in every time.

Keep exploring

More in AI Research