How to transcribe interviews

Qualitative transcription isn’t a one-size-fits-all task. The level of detail—what gets written down, what gets omitted, and how speech is cleaned up—should be chosen based on the study’s goals and the kind of analysis planned, because different research questions require different kinds of evidence. Attempts to create universal transcription rules have repeatedly failed, largely because qualitative research varies widely, and transcription choices can change what later arguments seem to support.

For most qualitative work that focuses on content (participants’ experiences, views, beliefs, and narratives), transcription can stay relatively readable and “clean.” In these studies, the transcript primarily functions as a record to help researchers remember and understand what was said so they can code and analyze themes, grounded theory categories, or phenomenological meaning. That typically means nonverbal cues—stutters, “ums,” pauses, and other delivery details—are not essential. Still, researchers should not ignore them automatically: if something stands out as analytically relevant—such as an unusually long pause or visible anxiety—then those cues should be indicated consistently, for example using brackets.

The main exception is discourse analysis, where the research question targets how meaning is constructed through language and interaction, not just what participants report. In discourse analysis, pauses, hesitations, stutters, and other interactional signals can be part of the evidence. That pushes transcription toward a much more fine-grained record, sometimes including line-by-line detail, because the analysis depends on timing, phrasing, and the structure of talk.

To justify transcription decisions later, the guidance centers on a practical continuum between two extremes: “naturalism” and “denaturalism.” Naturalism treats speech as raw data, preserving every hesitancy and filler word so the transcript mirrors natural conversation. Denaturalism, by contrast, cleans up speech so it reads smoothly and focuses on intelligible content. Most researchers land somewhere in the middle—leaning toward cleaned-up transcripts for efficiency—while selectively retaining or marking details when they carry interpretive weight.

A workplace example illustrates the stakes. In a cleaned-up transcript, a colleague’s comment about having a partner can appear friendly and straightforward. Under a more naturalistic transcription approach, the same exchange can include hesitation, awkward laughter, and nervous glances—signals that shift the perceived meaning from casual to uncomfortable or even “creepy.” The key takeaway is that transcription choices shape context and interpretation, so researchers should use judgment grounded in their study aims.

Ultimately, transcription is framed as a decision process: use common sense as the starting point, then align the transcript’s level of detail with the analysis method and the evidence needed to answer the research question. Consistency in how cues are marked matters, but there is no universal “correct” approach—only a defensible one tailored to the study.