Research operations workshop / 5 MIN READ
Check the transcript before a misheard word becomes a finding
Verify meaning, speaker attribution and uncertainty against the permitted recording.
AI-assisted research and analysis for the SNS Creative Fieldbook
Sources checked 2026-11-06
A transcript is a working representation of a conversation. Small errors in negation, product terms or speaker identity can change an interpretation even when the surrounding text reads fluently.
What you will leave with: A transcript verification record with corrections, unresolved passages and a clear release status.
Use this when
- Automatic transcripts are used for research summaries or quote retrieval.
- Several speakers or specialist terms make transcription uncertain.
This is not permission to upload recordings to an unapproved service or to rewrite a participant’s speech for a stronger story.
Agree what the transcript must preserve
University of Bath guidance asks researchers to decide the required transcription detail and how accuracy will be checked. Define the transcript’s intended use before judging whether it is adequate.
A thematic review may need different detail from research about conversational interaction. Specify treatment of pauses, overlap, incomplete phrases and inaudible material where relevant. Keep a summary distinct from a transcript. Confirm that the recording and any transcription provider fall within the approved research and data-handling arrangements before making a new copy or upload.
Verify passages that carry consequential meaning
Compare the text with the permitted recording, paying attention to product names, numbers, qualifications and negation. Check enough surrounding conversation to understand who or what a pronoun refers to. A fluent sentence can still be a poor representation of what was said.
Use the study’s quality procedure to decide the scope of checking. Any excerpt used to support a consequential claim needs direct verification; a spot check does not certify the whole file. If a pattern of errors appears, expand the review or hold the transcript from use. Do not advertise a universal accuracy percentage from an informal check.
Correct transparently and leave uncertainty visible
Keep the original machine or supplier output separate from the verified working version according to the approved retention process. Record meaningful corrections and identify the recording location that supports them. Use neutral uncertainty markers when the audio cannot resolve a word.
Check speaker labels, especially after interruptions or when several people share a microphone. Do not assign a remark to the participant merely because it sounds like customer feedback; it may be the facilitator’s suggestion. Remove identifying details from the authorized working copy where required without losing the protected link needed for accountable verification.
Release a usable record with limits
Mark the transcript’s status, checking scope and unresolved passages. Let analysts know which material must not be quoted as exact speech. Preserve the distinction between spoken wording, a translation and an editorial paraphrase. Readers should be able to tell what kind of evidence they are using.
If a corrected passage already informed a finding, review that finding and notify the appropriate collaborators. A file correction alone does not repair a slide or decision built on the earlier wording. Close the verification task when dependent uses have been checked and records are handled under the study’s retention and access rules.
Make the reasoning concrete.
A fictional research transcript says a buyer “would renew without approval.” Listening to the authorized recording reveals “wouldn’t renew without approval,” and a later paragraph is assigned to the wrong speaker.
Use the fluent transcript
Opening: Quote the sentence in a finding about easy renewals.
Hypothesis: The analysis can proceed quickly.
Risk: A missing negation reverses the participant’s meaning.
Verify and trace the correction
Opening: Correct the passage, check the speaker label and review the affected finding.
Hypothesis: The team can repair both the evidence and its downstream use.
Risk: Other consequential passages may need checking before release.
The researcher holds the affected finding, updates the verified version and marks an unrelated unclear phrase as unresolved rather than guessing. The report uses only wording whose meaning has been checked.
Take the questions into your next brief.
Use these prompts in your own notes. Nothing is submitted from this page.
Purpose
What level of detail must this transcript preserve?
Permission
Where may the recording be processed and stored?
Check
Which passages and speaker labels need verification?
Uncertainty
How are unresolved words or overlaps marked?
Dependency
Which findings or excerpts rely on corrected text?
Watch for the wrong conclusion.
- Trusting fluent text without listening to consequential passages.
- Silently guessing an unclear product term.
- Fixing the file while leaving a misleading quote in circulation.
Your next action
Verify one transcript passage that supports an important finding and record both the check and any downstream correction.
Sources & scope
The workflow, worksheets and hypothetical examples are original editorial analysis. Sources support only the identified principles. No measured business results or named human reviewer are claimed.
- University of Bath Library: Working with data, Transcription →Primary university research guidance · Supports: Transcript preparation requires decisions about detail and accuracy checks; the verification log and example are original editorial analysis.