Why the level matters
The same recording can produce very different transcripts. The verbatim level decides which features of speech stay in and which come out. It affects how the transcript reads, how long it takes to produce and what it can show later.
The terms are not standardised. One UK provider describes intelligent verbatim as removing "ums, errs and repetitions and false starts". Another describes its clean style as correcting grammar and restructuring sentences. Agree the rules in writing before work starts.
One answer at each level
The lines below are an invented answer from an investigation interview. Each version comes from the same audio.
| Level | How the line appears |
|---|---|
| Strict verbatim | So, um, the- the meeting was, it was the Tuesday. No, the Wednesday. He, er... (pause) he come in (laughs) and said they was all being moved. (door closes) |
| Full verbatim | So, um, the- the meeting was, it was the Tuesday. No, the Wednesday. He, er, he come in (laughs) and said they was all being moved. |
| Intelligent verbatim | It was the Tuesday. No, the Wednesday. He come in and said they was all being moved. |
| Edited, for comparison | It was Wednesday when he came in and said they were all being moved. |
The edited line reads well but loses information. It hides the speaker’s change from Tuesday to Wednesday, and it alters the speaker’s grammar. In an investigation, either detail could matter.
What each level keeps and removes
Intelligent verbatim removes fillers, false starts, stutters and repetitions that add no meaning. It keeps self-corrections that change the meaning, as in the example. In legal and HR work it normally keeps the speaker’s own grammar and dialect.
Full verbatim keeps every word spoken, including fillers, false starts and repetitions. It usually notes sounds that carry meaning, such as laughter.
Strict verbatim keeps everything in full verbatim and adds pauses, short responses such as "mm-hm", background sounds and overlapping speech. Each feature is marked using an agreed convention.
| Feature | Intelligent | Full | Strict |
|---|---|---|---|
| Fillers such as "um" and "er" | Removed | Kept | Kept |
| False starts and repetitions | Removed unless meaningful | Kept | Kept |
| Stutters and partial words | Removed | Kept | Kept |
| Laughter and similar sounds | Noted where relevant | Noted | Noted |
| Pauses, "mm-hm" and background sounds | Left out | Usually left out | Marked |
| Speaker’s grammar and dialect | Usually kept as spoken | Kept | Kept |
Even these lines vary between projects. An example set of research instructions from the UK Data Service keeps fillers and false starts. It leaves out the interviewer’s short responses, so the participant’s answer reads without breaks.
Which level suits which work
For most HR investigations, disciplinary hearings and grievance meetings, intelligent verbatim is enough. It reads clearly and keeps the speaker’s words. Choose full or strict verbatim where hesitation, interruption or exact phrasing may be disputed, such as an account of a heated exchange.
Legal readers often care whether a witness hesitated or changed course. Court transcript conventions reflect this. One official court transcription provider’s guide uses a dash where a speaker makes "a false start or changes tack".
Research needs depend on the method. The UK Data Service describes a denaturalised approach, closer to written language, for thematic projects. It describes a naturalised approach, closer to speech, for conversation analysis. Detailed discourse work may need strict verbatim or Jefferson notation, which uses symbols for timing, pitch and overlap.
Fuller levels take longer to produce and usually cost more. Published UK price lists often add a supplement per audio minute for full or strict verbatim.
Agreeing the style
Before work starts, agree these points in writing:
- the verbatim level, and whether grammar and dialect stay as spoken
- how self-corrections, interruptions and overlapping speech are shown
- how speakers are labelled, by name, role or code
- whether timecodes are needed, and how often
- how unclear words, redactions and anonymised names are marked
Glossary of transcript terms
- Speaker attribution: labelling each turn with who spoke, by name, role or code, such as "Interviewer" or "Witness A". Agree how to label a voice that cannot be identified.
- Timecode: a time reference from the recording, such as [00:14:32]. Timecodes can appear at set intervals, at each change of speaker or beside unclear passages.
- Inaudible marker: a note that words could not be heard, such as [inaudible 00:14:32]. A related marker shows a best guess, such as [unclear: Thursday?]. Court transcripts commonly use (inaudible) and (?).
- Crosstalk: two or more people speaking at once. It is marked [crosstalk], or the overlapping words are shown where they can be separated.
- Q&A layout: questions and answers set out as alternating lines marked Q and A, often with line numbers. It suits interviews and witness accounts, and line numbers make corrections easy to locate.
- Redaction marker: a placeholder showing that information has been removed, such as [name removed] or [Colleague 1]. The UK Data Service advises placing anonymisation edits in square brackets.
- Sic: a note, usually written (sic), that an error is the speaker’s own and has been kept as spoken.
- Non-verbal annotation: a short description of a sound or action, such as (laughs) or (pause).
You can see speaker labels, numbered lines and a question and answer layout in the sample records.
Sources
- UK Data Service: Transcription
- UK Data Service: Example transcription instructions (PDF)
- eScribers: Guide to symbols used in a transcript (PDF)
- University Transcriptions: Transcription editing levels
- CareScribe: Guide to verbatim transcriptions
General information. Prices and guidance change; check the sources above for the current position.