What a raw interview transcript actually looks like

A transcribed video interview arrives as continuous text with speaker labels. The raw output includes every hesitation, every unfinished sentence, every sidebar that happened before the real answer started. In a 45-minute interview, roughly a quarter of the text is preamble, filler, and off-topic exchanges that serve the live conversation but contribute nothing to the published piece.

The Q&A format imposes specific conventions on this material: questions are short and specific, answers are complete and self-contained, attribution is consistent, and the sequence has an editorial logic (not necessarily chronological). The raw transcript meets none of these conventions automatically. All of them require deliberate editing.

Step 1: Speaker diarisation and attribution cleanup

Attribution before editing. Always. Get speakers right. Wrong attribution is the worst error in a published Q&A.

Misattributing a statement to the wrong speaker in a published Q&A is categorically different from leaving in a verbal tic. In some contexts, attributing a damaging statement to the wrong person can constitute defamation, whereas a filler word that survived the edit is merely an editorial lapse with no legal consequence beyond looking slightly informal to readers familiar with edited interview conventions. The SPJ Code of Ethics covers the attribution and accuracy standards that apply to published interview content in journalism contexts.

If the transcript was generated by an AI tool, speaker labels are usually generic: Speaker 1, Speaker 2. Rename them before doing any other edit. Replace with the actual format you will use in publication: Q and A, or name-based attribution. This is mechanical but doing it first prevents confusion during content editing.

Check speaker assignment accuracy throughout. AI diarisation errors cluster around: moments where the speaker changes mid-sentence, long pauses that reset the diarisation, and sections where both voices are similar in pitch or cadence. Read the raw transcript against your memory of the interview and correct any mislabelled sections before starting the content edit.

Step 2: Select and rewrite the questions

Tight questions. One topic each. Not three joined by "and also." Written Q&A questions do not meander. Rewrite every question before touching the answers.

The questions you actually asked in the interview are rarely the questions you want in the published Q&A. Interview questions in the moment tend to be long, contextual, and shaped by what the previous answer was. In the published format, each question must work without the conversational context.

Rewrite each question to be:

Specific. “Tell me about your approach to this” becomes “How do you decide which projects to take on?”
Concise. Cut the preamble (“You mentioned earlier, and I thought this was really interesting, that you...”). Keep the question.
Coherent without the previous answer. If a question references “what you just said about X,” rewrite it to state X explicitly or restructure the answer to eliminate the reference.

Step 3: Edit the answers for length and clarity

Answers are always too long. Every interview. The editing rule: cut to the point where the answer first becomes complete. Everything after that is repetition. Cut it.

Answers are too long. Every time. Without exception. The spoken version was designed to fill silence and build rapport. The written version has neither of those requirements. Cut to the point.

Interview answers in transcripts are usually 2-4x longer than they need to be in print. The editing process is aggressive but specific: you are not changing what the subject said, you are removing the parts that were said for spoken reasons (building to a point, repeating for emphasis, buying time to think).

Cut first. Read after. The key distinction: if you read through an answer before cutting, the length feels justified because you just spent time engaging with the content — this is a well-documented cognitive bias that makes editors reluctant to cut material they have just processed, even when that material adds nothing for readers. Cut first. The reading pass confirms what survived.

Cut first. Read after. Two passes. Done.

The ethical line is clear: you can cut, you cannot add. You can rearrange sentences within an answer if it improves clarity without changing meaning. You cannot move content to misrepresent what someone said. I've worked on interview Q&As where the subject later objected to cuts — the rule I follow is to never cut anything that reverses the overall impression, even if the individual section seemed redundant. That standard keeps the edit defensible. material from one answer to another. The standard notation in published Q&As is a brief editor's note stating responses have been edited for length and clarity. This signals transparency without itemising every cut.

Multi-speaker interviews: the hardest case. Three participants. Overlapping answers. Attribution errors that are invisible until a subject objects. Verify before publishing. Every attribution. No exceptions.

For multi-speaker interviews or complex research citation requirements, the guide to using transcripts in research covers citation standards and attribution conventions in more detail.

Step 4: Sequence decisions

Published Q&As rarely run in chronological interview order. The natural conversation arc of an interview — warm-up questions, the substantive middle, the closing wind-down — does not make the best reading order. Consider resequencing so the most substantive exchanges come early. Readers drop off; front-loading the strongest material keeps more of them through to the end.

Resequencing feels wrong at first. It should not. The original interview order served the conversation. A different order serves the published reader. Those are different jobs. Do not confuse faithfulness to the event with usefulness to the audience.

One common approach: open with a question that establishes the subject's credibility or main claim, follow with the technical or detailed content, close with a forward-looking or opinion question. This is not how interviews happen naturally, but it is how most readers prefer to read them.

Multi-subject interview transcripts: managing speaker attribution

Three speakers: hard. Four speakers with crosstalk: very hard. The attribution pass is not optional for panel transcripts. AI assigns words to the wrong speaker more often than most editors expect the first time they encounter it.

Panel discussions and multi-subject interviews are the hardest interview transcripts to work with. AI transcription for three or more speakers produces attribution errors at a higher rate than single-speaker or two-speaker content — particularly when speakers have similar voice profiles or frequently interrupt each other. Before editing, do a first pass specifically to check attribution: read the transcript while listening to the audio at 1.5x speed, correcting only speaker labels without touching the content.

In the Q&A format, multi-subject content often reads better if reorganised by subject: all responses from Speaker A on topic 1, then all responses from Speaker B on topic 1, rather than a chronological record of the conversation. This is a significant editorial decision that changes the nature of the document from a conversation record to a curated set of positions. Both formats are legitimate; the choice should be transparent in the published piece.

For academic or journalistic multi-speaker interviews that require verbatim accuracy, the standard is each speaker sign off on their attributed quotes before publication. This is standard practice at major publications and avoids misattribution disputes. Sending the specific quotes attributed to each person — not the full transcript — for confirmation is the practical approach. The review is faster for the subject and the confirmation is specific enough to be meaningful.

For interview-based research, sending subjects only the quotes attributed to them — rather than the full transcript — results in much faster confirmation turnaround. The subject reviews two or three specific passages rather than reading 5,000 words. The confirmation is also more credible: they are confirming specific words, not a full document they may or may not have read carefully. The attribution confirmation email is most effective when it is short: the quoted passage, a specific question asking for confirmation, and nothing else.

From interview transcript to podcast show notes

A video interview that's also published as a podcast episode needs two written deliverables from the same transcript: show notes and an article version. Show notes are a compressed summary — bullet points covering the main topics, key quotes, and relevant links, running to 200–400 words. The article is the full edited Q&A at publication length.

The right sequence is to complete the full Q&A edit before producing show notes. Show notes should compress the edited piece, not the raw transcript — this ensures both documents stay consistent because they came from the same edited source.

The efficient path: do the full Q&A edit first (the longer version), then compress the edited Q&A into show note bullet points. The show notes become an executive summary of the article, and both documents stay consistent because they came from the same source material. This is faster than producing both documents independently from the raw transcript.

Timestamped show notes — noting where each major topic starts — are more useful than unordered bullet points for listeners who want to jump to a specific section. Timestamps are the whole point. Without them, show notes are just a list. With them, they become navigation. If you have the SRT file, the timestamps are already there. Copy the segment timestamps for the start of each major topic into the show notes as (12:45) links. This is the main practical reason to download SRT alongside TXT even for a transcript you plan to edit into written content — the timing data disappears in TXT and can't be recovered without re-running the transcription.

The accuracy threshold that matters for Q&A

Attribution errors are the priority. Not homophones. Not filler words. Which speaker said what — that is what must be verified before any quote reaches publication.

AI transcription accuracy for interview audio typically runs at 92-96% for clear, single-speaker sections and drops to 80-85% for two-speaker exchanges with any overlap. At 95% accuracy, a 5,000-word interview transcript contains roughly 250 word-level errors. Most are small (wrong homophone, missed article), but attribution errors — text assigned to the wrong speaker — require careful human review before publication.

Frequently Asked Questions

How do I handle overlapping speech in an interview transcript?

AI transcription tools handle overlapping speech by assigning the text to whoever was speaking louder, or by splitting it imperfectly across speakers. In the edit, read the overlapping section in context: whose point was it advancing? Assign accordingly. If the overlap contains a genuine interjection ('Right', 'Exactly'), you can either keep it in brackets as [agrees] or delete it entirely. The Q&A reader does not need every vocal response recorded.

Should I edit the questions in an interview transcript?

Yes. Interview questions spoken in real time are often long, rambling, and imprecise. The published Q&A version should have tight, specific questions. The rule: the question should state exactly what the answer is about, nothing more. Rewrite long multi-part questions as two separate questions if needed.

Is it ethical to edit interview responses in a published Q&A?

Editing for clarity, concision, and to remove verbal filler is standard journalistic and editorial practice. What is not acceptable: changing the meaning of a response, removing material that would reverse the impression given, or adding content not in the original. Most publications note somewhere that responses have been edited for length and clarity.

How do I attribute speakers in a Q&A format?

Standard format: the interviewer is identified by role (Q: or Interviewer:) or initials. The subject is identified by name and title on first appearance, then by name only or initials. Long interviews with one subject often use Q: and A: throughout, which is cleaner than repeating names. Multi-subject interviews need consistent name attribution per answer.

What is the right length for a published interview Q&A?

There is no universal rule, but most published interview Q&As run 800–2,500 words. Below 800 words feels superficial for a full interview — not enough exchange to develop any subject meaningfully. Above 2,500 words requires extraordinary subject matter or a well-known subject to hold reader attention for the full length. A 45-minute interview transcript of 6,000–7,000 words usually edits to 1,500–2,000 words for publication — roughly 70% is cut, which is normal and appropriate. The goal of editing is not to preserve a record of everything said but to produce the version of the conversation most useful to readers who were not present.

How do I handle a subject who wants to review the transcript before publication?

Transcript review is a courtesy, not an obligation — unless you explicitly agreed to it. If a subject requests changes, the editorial standard is to accommodate corrections of factual error (they misspoke and want to clarify), but not retractions of accurate statements they now regret. Offer the subject the quoted passages attributed to them, not the full transcript, to focus the review on what will actually appear in print.

Video Interview Transcript to Q&A Format: Editing Guide