An interview can run for an hour and still hinge on six seconds: the moment a source hesitates, corrects themselves, or says the thing they did not mean to say out loud. Transcription is where those six seconds either survive intact or quietly disappear. Most errors are not dramatic. A dropped qualifier, an unattributed quote, a timestamp pointing at the wrong minute — small slippages that stack up until a fact-checker or an editor starts asking questions. The fix is rarely more effort. It is a better method, applied consistently.
1. Treating "verbatim" as a single standard
Verbatim means different things to different desks. Full verbatim captures every stumble, repetition and filler word. Clean verbatim removes the noise while preserving meaning. Broadcast or court-style transcripts may keep hesitations but tighten grammar. Pick your standard before you start, write it at the top of the file, and stay with it.
Here is the difference in practice:
Raw: "So, I mean, we didn't — we hadn't actually signed anything at that point, if I remember rightly."
Clean: "We hadn't actually signed anything at that point, if I remember rightly."
Note what stayed: the hedge "if I remember rightly". That phrase tells you how certain the source is, and it is not yours to delete. The same goes for "I think", "probably" and "as far as I know". Strip them and you have not tidied a sentence; you have changed a claim.
Keep, even in a clean transcript
- Hedges and qualifiers that signal uncertainty
- Self-corrections that show a change of position
- Numbers, dates and sums exactly as spoken, with a flag if they sound implausible
- Emotional markers where they matter — a pause, a laugh, a raised voice
2. Guessing at unclear audio instead of marking it
Names, place names, acronyms and job titles are where transcripts go wrong most often. You hear something that could be "Councillor Bream" or "Councillor Breen", and the temptation is to pick one and keep moving. Resist it. A guessed surname becomes a printed correction, and a misheard job title can turn a volunteer into a director.
Use a clear marker and a timestamp instead: [unclear 00:14:22]. If a whole phrase is lost, write [inaudible 00:14:22 – 00:14:29]. Then do the work:
- Listen three times, once at reduced speed.
- Check the surrounding context for clues.
- Cross-reference against an agenda, a website or earlier correspondence.
- Email the source a short list of queries, with timestamps, rather than a whole file.
A transcript with four honest gaps is more useful than one with four confident inventions.
3. Losing speaker labels and turn boundaries
With two or more voices, ambiguity creeps in fast. Overlapping speech is the usual culprit: the interviewer says "and what happened next?" while the subject is still finishing a sentence, and a careless transcript hands one person's words to the other. Label every turn, consistently, from the first minute. Use the full name on first appearance, then a clear short form.
Do not merge two answers into one quote because they read more smoothly that way. If a source pauses for eight seconds, then adds a crucial caveat, that caveat belongs in a separate paragraph. Mark overlaps as [overlapping] and note when a speaker is reading from a document. Anyone quoting the transcript later will thank you for it.
4. Inconsistent style, timestamps and formatting
Half a transcript in -ise spellings and half in -ize looks careless, even when every word is accurate. So does a timestamp that changes from 00:04:12 to 4.12 to four minutes in. Agree the details at the start and keep a style sheet for the job: spelling conventions, how numbers and dates appear, whether contractions are retained, how long pauses and laughter are shown.
Before you call it finished
- Search the file for [?], [unclear] and [inaudible]. Deal with each one or leave it visible.
- Check every proper noun against a primary source, not your memory.
- Read each quotation against the audio, word for word.
- Confirm speaker labels have not drifted.
- Verify figures, dates and titles, and flag anything you could not confirm.
5. Skipping the listening passes
Typing a transcript and then editing it on screen alone is how small errors survive. Reading aloud exposes dropped negatives — "I did not agree" losing its "not" is a serious mistake in any context, and a catastrophic one in a dispute. Play the audio once more while following the text with your cursor. It is slower than skim-reading. It is also the only reliable way to catch a sentence that is grammatically perfect and factually wrong.
Two further habits belong here. First, protect the material: keep the original recording, store files with a consistent naming convention, and do not circulate drafts more widely than necessary, particularly where a source has spoken in confidence. Second, if a transcript may be used in legal or formal proceedings, check the required format and standard with a qualified professional before you rely on it. General good practice is not the same as legal compliance.
A repeatable workflow that keeps errors down
Good transcripts come from rhythm rather than heroics. Listen to the whole interview once without typing. Note names, jargon, timestamps and anything you will need to query. Then transcribe in focused blocks of twenty to thirty minutes, marking unclear passages as you go instead of stopping to solve them. Take a short break between blocks; attention is the limiting factor, not typing speed.
After the draft, do three passes in this order: a clean-up pass for style and readability, a verification pass against the audio, and a final read-aloud for sense. Send source queries as a numbered list with timestamps so they are quick to answer. Keep the marked-up draft until the piece is published, then archive it with the audio.
Accuracy is not a flourish you add at the end. It is the thing that makes the rest of the work usable. Get the method right and the transcript stops being a chore in the background and becomes what it should be: the solid ground beneath every quote you publish.
Photo: theglassdesk / Pixabay
