Wie lange dauert es, ein Interview zu transkribieren?

Von Hand gilt als Faustregel das Fünf- bis Zehnfache der Aufnahmedauer. Mit KI-Vortranskription und anschließender Korrektur am Ton liegt der Aufwand meist beim Ein- bis Zweifachen.

Welche Transkriptionsart brauche ich?

Für Marktforschung, Journalismus und Berichte reicht die sinngemäße Transkription. Wortwörtlich wird transkribiert, wenn auch das Wie zählt, etwa in der Gesprächsanalyse oder bei juristischer Verwendung.

Ist KI-Transkription gut genug?

Als Rohfassung ja. Bei mehreren Sprechern, Dialekt, Fachbegriffen und Hintergrundgeräuschen entstehen Fehler, die nur am Ton auffallen. Deshalb folgt auf die KI-Fassung eine Korrektur durch eine Fachkraft.

Was muss ich beim Datenschutz beachten?

Einwilligung vor der Aufnahme einholen, personenbezogene Angaben im Transkript pseudonymisieren, Speicherort und Löschfrist festlegen. Wer extern transkribieren lässt, schließt einen Auftragsverarbeitungsvertrag.

Kann ich Interviews auch übersetzen lassen?

Ja. Üblich ist, zuerst zu transkribieren und den Text anschließend zu übersetzen, weil dann Fachbegriffe und Sprecherzuordnung feststehen.

Guide · For business

Transcribing interviews: types, rules and process

Which type of transcription suits which purpose, which rules to set before the first transcript and what the process looks like with AI pre-transcription. With a checklist for the recording and the data protection requirements.

9-minute read
Updated 16 October 2026

In short

There are three types of transcription: verbatim, clean verbatim and edited. Choose by purpose, not by preference.
Transcribing by hand takes about five to ten times the length of the recording. With AI pre-transcription and correction, that falls considerably.
The biggest gains come before the interview: a quiet room, the right microphone and a test recording.
Set the transcription rules before the first transcript, or you will have to standardise everything afterwards.
Consent before recording, pseudonymisation in the transcript and a data processing agreement if transcription is outsourced.

The three types of transcription

Transcription means turning spoken language into text. How exactly that happens depends on the purpose. If only the content matters, you transcribe clean verbatim. If how something is said matters too, including pauses, slips of the tongue and emphasis, you transcribe verbatim. If the text is going to be published, it is edited as well.

Type	What is included	Suitable for
Verbatim	Every word, plus filler words, repetitions, pauses and non-verbal sounds	Qualitative research, conversation analysis, legal purposes
Clean verbatim	The content without filler words, slips of the tongue and throat-clearing	Market research, journalism, reports: the most common case
Edited	Grammar also smoothed and sentences put in order	Content, interviews for publication, subtitles

Before recording: checklist

The quality of the recording determines how much work follows. Half an hour of preparation saves hours on the transcript later, and some problems can’t be fixed at all afterwards.

Tool · Checklist

Preparing the recording and transcript

Quiet room chosenCarpets and curtains dampen sound; switch off air conditioning and fans beforehand
Microphone suited to the situationLapel microphones for one-to-one interviews, a boundary microphone for groups, a smartphone only in a quiet room
Test recording made and played backTwo minutes are enough to detect echo, hum or a recording level that is too low
Battery, storage and backup device checkedA second recording on a phone is the cheapest insurance
Consent obtainedBefore you start, in writing, covering the recording, processing and planned use
Speakers asked to say their names at the startMakes assigning speakers much easier later
Transcription rules agreedType of transcription, timestamps, how to handle inaudible passages
Filing and backup arrangedFile names, storage location, access rights and deletion period

Setting transcription rules

Transcription rules define how speech is put into writing. Set them before the first transcript, so that all the files in a project look the same and can be compared later. These nine points are enough for most projects.

Element	Recommendation
Speaker labels	Fixed abbreviations, such as I for the interviewer and R1, R2 for respondents
Timestamps	Every three to five minutes or at every change of speaker
Inaudible passages	Mark consistently, such as [inaudible] with the time
Uncertain words	In brackets with a question mark, such as (supply chain?)
Pauses	A consistent symbol, such as (…) for pauses of three seconds or more
Non-verbal sounds	Only if they matter for the analysis, such as [laughs], [phone rings]
Dialect and colloquial speech	Decide whether to keep it as spoken or smooth it into standard language
Numbers and abbreviations	Write them out or use digits, and stick to one approach
Anonymisation	Which details are replaced and with what, such as [Company A], [City]

Established rule systems are common in academic research. In German-language research, these include the Dresing and Pehl system for content analysis and GAT for conversation analysis. In English-language conversation analysis, Jefferson notation is widely used. If your university or client specifies a system, it takes precedence.

Process and effort

The usual approach today is AI pre-transcription, followed by correction against the audio. The draft takes minutes, but the checking is still done by hand.

ReviewListen to the recording once and mark difficult passages
Pre-transcriptionAI produces the draft with timestamps
Correction against the audioCheck section by section, correct speakers and technical terms
FormattingApply the rules, anonymise
Final checkSpot checks against the audio, completeness, filing

Five steps from audio to finished transcript.

Method	Time required	Note
Typing by hand	About five to ten times the length of the recording	So a good working day for a one-hour interview
AI transcription	A few minutes	The result is a draft, not the finished transcript
AI plus correction	About one to two times the length of the recording	The usual approach if the transcript needs to be quotable

Where AI transcription reaches its limits

Situation	What happens
Several people speaking at once	Changes of speaker and overlaps are attributed incorrectly
Dialect and strong accents	Words are guessed, often plausibly but still wrongly
Technical terms and proper names	Product names and abbreviations come out garbled; a glossary helps
Background noise	Traffic noise and room echo reduce accuracy considerably
Numbers and units	Often written down wrongly, especially in interviews with a lot of data

Data protection and anonymisation

Interview recordings are personal data, often even when no names are mentioned. Obtain consent before recording, covering the recording, processing and planned use. Decide on the storage location, access rights and deletion period before the first file is created.

In the transcript, personal details are pseudonymised: names are replaced with codes, places and companies are described in general terms and exact ages are converted into age ranges. Then check whether people could still be identified from details, such as an unusual job title in a small company.

If you outsource transcription or use a cloud tool, you need a data processing agreement under the GDPR and should know where the servers are located. At tolingo, transcription runs in processes certified to ISO/IEC 27001. This is guidance, not legal advice.

Common mistakes

Most problems arise before transcription starts: missing rules, a poor recording or unresolved data protection questions.

Avoid	Better	Why
Transcribing first, setting rules later	Set the rules before the first transcript	Otherwise half the file has to be standardised afterwards.
Transcribing verbatim when clean verbatim is enough	Match the type of transcription to the purpose	Verbatim takes many times longer and is harder to read.
No timestamps	A timestamp every three to five minutes	Without them, every check turns into a search.
Leaving real names in the transcript	Pseudonymise, such as [Respondent 1], [Company A]	Usually a requirement for research and publication.
Any AI tool for sensitive recordings	A provider with a data processing agreement	Without an agreement and a known server location, the processing lacks a sound legal footing.
Passing on the AI draft unchecked	Check it against the audio	Otherwise errors in figures and names only come to light once they are quoted.

FAQ

How long does it take to transcribe an interview?

By hand, the rule of thumb is five to ten times the length of the recording. With AI pre-transcription followed by correction against the audio, it usually takes one to two times as long as the recording.

Which type of transcription do I need?

For market research, journalism and reports, clean verbatim is enough. Verbatim transcription is used when how something is said also matters, for example in conversation analysis or for legal use.

Is AI transcription good enough?

As a draft, yes. With several speakers, dialect, technical terms and background noise, errors creep in that only show up when you check against the audio. That is why the AI version is followed by a correction by a professional.

What do I need to consider for data protection?

Obtain consent before recording, pseudonymise personal details in the transcript and set the storage location and deletion period. If you outsource transcription, sign a data processing agreement.

Can I have interviews translated as well?

Yes. The usual approach is to transcribe first and then translate the text, because the technical terms and speaker attribution are settled by then.