A recorded interview becomes useful when it turns into evidence attached to skills. It becomes dangerous when it turns into a smooth summary nobody checks.
Permission first
Recording needs the candidate's clear agreement, and in some places everyone else's too. Ask before the interview, not in the first minute of it. Consent asked for while someone is already on camera is not really free.
Tell them what is recorded, who can see it, how long you keep it, what it is used for, and that saying no costs them nothing. Then make saying no actually cost nothing, which means having a process that works without the recording.
Evidence, not summaries
Most transcript tools produce a summary by default. That is the least useful form the information can take. A summary is squashed down, has no source attached, and cannot be checked without reading the thing it replaced.
What you want instead is extraction: for each skill on your scorecard, the passages where it came up, quoted.
| Not this | This |
|---|---|
| "Showed strong stakeholder management" | "Described handling a disagreement with finance over budget, quote: …" |
The left one is a conclusion. Whose? Based on what? The right one is evidence — checkable, and the person reading it draws the conclusion.
The rule: the software collects, the human judges. Doing it the other way round is how a hiring decision ends up resting on a sentence nobody verified.
Two different ways it goes wrong
- Mishearing. Transcription gets worse with accents, people talking over each other, poor audio and technical words. Usually the errors are obvious nonsense — but not always, and a believable mistake in a number or a company name is worse than an obvious one.
- Inventing. Ask AI to summarise an interview and it will produce the responsibilities that are typical for that job, rather than the ones actually mentioned. Same problem as Module 1, Lesson 3.
The protection is the same as everywhere else in this course: ask for quotes with timestamps, and check three of them against the recording. If your tool cannot give you a timestamped quote, it is giving you an impression.
Turning it into scores
- Pull out quoted passages for each skill.
- Read your written description of what a 3 looks like.
- Give a score, with one quote as the reason.
- Where there is no evidence, write "no evidence" rather than a middle score.
That last step is the one people skip. A 2 given because nothing came up is not the same as a 2 given because the answer was weak. Treating them the same destroys the meaning of the score — in the discussion, and in the record you may later have to defend.
What never goes in the scorecard
Anything from the transcript that is not about a skill. Personal circumstances, health, family, accent, or anything else mentioned in passing. If your tool surfaces that, it is a reason to look at what your tool is extracting.