Interview and Transcript Analysis

How do I analyse an interview transcript properly?

Assess the transcript's quality, read it whole without annotating, map the conversation's phases, separate what the moderator introduced from what the participant brought, then code for meaning with the position and prompt level attached to every passage. Only then write the account. Coding before reading produces decisions you cannot see, and coding without positions produces statements you cannot weigh.

How do I tell the participant's own view from something the moderator suggested?

Build a ledger of every concept, word or framing the moderator put into the conversation, note where it entered, and track each one forward. When a participant uses a moderator's term thirty minutes later it reads exactly like a spontaneous insight and is routinely reported as one. If the guide asked "how important is convenience?", a later statement that convenience matters is a finding about the guide.

What is the difference between volunteered and prompted material?

Volunteered means the participant raised it with no prompt on the topic, and it is the strongest evidence of what actually matters to them. Elicited means an open question on the topic without the content supplied. Prompted means a specific probe naming the thing. Confirmed means agreeing with something the moderator asserted, which is the weakest of the four and is frequently just acquiescence.

Why do participants give such neat explanations of messy decisions?

Because they are reconstructing. An account of a past decision is assembled afterwards from memory, from what they have since learned, from what makes them look competent, and from the version they have told colleagues. A clean, ordered, well-reasoned account of a messy choice is a signal, not a finding. Report it as their current explanation, which is genuinely useful, rather than as the decision process, which it is not.

What should I do when a participant contradicts themselves?

Record both statements with their positions and what prompted each, then characterise the contradiction rather than resolving it. It usually means one of five things: the statements are about different contexts, one is the socially acceptable answer and the other is the lived one, they genuinely hold both, the topic shifted, or the moderator reframed and they followed. Choosing the more coherent version deletes the evidence.

Can I analyse a single interview, or do I need a full sample?

You can, and the output is a case, not a theme. One interview supports strong claims about a mechanism and no claims at all about a population. What breaks is counting: "five of eight participants" on a base of eight is a description of those eight and will be quoted as a proportion the moment it reaches a slide.

How is this different from thematic analysis?

This skill works within and across a small number of interviews at depth, keeping context and sequence, and produces case records. Thematic analysis builds and tests themes across a full dataset. Run thematic analysis on three transcripts and your themes are really three participants; run this on thirty and you get thirty case records and no analysis.

How do I check transcript quality?

Read for defects before reading for content: missing or swapped speaker labels, the proportion marked inaudible, sentences that do not parse, and the characteristic errors of automated transcription, which are confident mis-hearings that produce plausible sentences rather than obvious gibberish. Where audio exists, spot-check the passages you would most want to quote, because those are the ones that will travel furthest if they are wrong.

Can I analyse a translated transcript?

Yes, with restrictions. Translation adds a layer of interpretation, and word choice in a translated transcript belongs to the translator, not the participant, so close reading of specific wording is not available. Idiom, deference, indirectness and humour are exactly where interpretation goes wrong. A reviewer with the relevant language and cultural context should read before findings are fixed.

How do I analyse an AI-moderated interview?

Audit the probing before trusting the content. A scripted prober introduces terms with perfect consistency and no awareness that it is doing so, so every participant produces content on every list item and almost none of it is volunteered. Check whether follow-ups responded to the participant's own words or to the script, and report the probing quality as a fieldwork finding.