# Keep the whole interview.

Canonical: https://axiomcx.dev/guides/interview-transcript/
Publisher: Axiom Cortex / TeamStation AI
Content reviewed: 2026-09-26

Provide the full interview with speaker attribution, question boundaries, and accurate technical wording. Check the transcript against the recording, keep meaningful hesitations and false starts, and mark unclear sections instead of guessing. A summary is not a substitute. No transcription tool guarantees perfect capture, so review the exported file before evaluation.

We use science to align tomorrow’s IT talent with the work ahead.

## Four steps to usable evidence.

1. Keep the authorized recording.
2. Export the full transcript.
3. Check words, speakers, and coverage.
4. Connect it to the role and questions.

Keep the source beside the transcript. A messy transcript can change what the evidence supports. Missing negation, a wrong technical term, or an answer credited to the wrong speaker can change what the interview supports.

## Use the recording you have.

### Google Meet

Use the meeting’s transcript where your Workspace edition and admin settings allow it. Keep the complete document, then check it against the recording.

[Google Meet transcript guide](https://support.google.com/meet/answer/12849897).

### Zoom

Check whether audio transcription is enabled for the recording. Preserve the transcript and original recording together.

[Zoom transcription guide](https://support.zoom.com/hc/en/article?id=zm_kb&sysparm_article=KB0064927).

### Microsoft Teams

Use the meeting transcript export available to an authorized participant. Check the speaker labels and any missing sections.

[Teams transcript guide](https://support.microsoft.com/en-us/teams/meetings/start-stop-and-download-live-transcripts-in-microsoft-teams).

Feature access, export formats, languages, and retention depend on the provider and your account settings. Check the official guide before the interview. Keep the transcript text, not just a meeting summary. Don’t let the app smooth away the detail.

## Hosted or under your control.

### TurboScribe

A hosted upload option for existing audio or video. Check the current free or paid file limit, enable the relevant speaker options, and review the full export.

[Check TurboScribe options](https://turboscribe.com/).

### Rev

Request full verbatim explicitly when fillers, false starts, and repetitions matter. Confirm timestamps, speakers, and the selected human or AI service.

[Read Rev’s verbatim guide](https://www.rev.com/resources/verbatim-transcription).

### Sonix

Check the documented verbatim option and timing export. Review overlapping speech, speaker attribution, and technical terms.

[Read Sonix’s verbatim guide](https://sonix.ai/verbatim-transcription).

### Local Whisper

Run transcription in an environment you control. Record the model and settings, retain the raw output, and check missing or invented words.

[Open the Whisper repository](https://github.com/openai/whisper).

These are preparation options, not verified Axiom integrations or guarantees of complete capture. Local execution avoids a hosted upload dependency; it does not make transcription error-free. Confirm permissions and data terms for whichever path you choose.

## Keep the detail that changes meaning.

- Full coverage: the opening baseline, every question, follow-ups, and the end of the interview.
- Attribution: distinguish candidate, interviewer, and other speakers.
- Exact meaning: check names, numbers, negation, code terms, and system boundaries.
- Source locations: keep reliable timestamps and question references where available.
- Uncertainty: mark inaudible or overlapping sections; do not fill them in with a guess.
- Speech detail: keep meaningful hesitations, repetitions, and false starts instead of smoothing them away.

[Download the transcript checklist](/guides/interview-transcript/checklist.md)

### A synthetic example: one missing word changes the finding

> Candidate: “We did not retry every request. We retried only idempotent operations.”

Synthetic teaching example. Not a real candidate.

A summary such as “Used retries” loses the safety condition. Preserve the words and source location so a reviewer can inspect the actual reasoning.

### Speech patterns need more than plain text

Ordinary punctuation does not establish a measured pause. A cleaned transcript cannot establish timing, tone, overlap, or whether an answer was read. Baseline-conditioned integrity review needs usable source recording, timing, speaker turns, and context.

- [Read the baseline and L2 review rules](/guides/interview-integrity/)

## Keep exploring

- [how transcript evidence is evaluated](/processing-engine/)
- [understand baseline integrity review](/guides/interview-integrity/)
- [evaluation prices and quantities](/pricing/)

[See the input workflow](/processing-engine/)