Transcription is the most useful thing AI has given a recruitment desk in a decade. It is also the clearest illustration of what a system cannot do, because everything that matters in a screening call happens in the gap between what was said and what was meant.
Ask a candidate why they are leaving, and time the pause.
A quick answer is usually a prepared one, which means it is the acceptable version. A pause before an ordinary question means they are deciding how much to tell you. That pause is the most valuable second in the call and it appears in a transcript as nothing at all.
Twenty years on a desk teaches you a short list of tells, none of which survive transcription.
The answer that is too complete. Someone who has rehearsed why they are leaving has usually rehearsed it because the real reason is harder to say.
The name they do not mention. A candidate who describes two years of work without once referring to their manager is telling you about their manager.
The enthusiasm that arrives late. When someone brightens twelve minutes in, at a detail you almost skipped, you have found what they actually want. That detail is now the centre of your pitch.
What they ask about. Candidates reveal their priorities in their questions far more reliably than in their answers.
A transcript is a complete record of the words and a poor record of the conversation.
Not in the screening call, which the recruiter attended. It bites three weeks later, when the placement is being decided by people who were not there.
A client reads a summary and sees a competent candidate. The consultant remembers someone who lit up talking about mentoring juniors, in a team that badly needs it, and who will take slightly less money for a role that offers it. If that never leaves the consultant's head, the client is choosing between two documents rather than two people.
This is not an argument against recording interviews. It is the opposite. A consultant typing notes is a consultant half-listening, and half-listening is exactly how you miss the pause.
Let the system produce the record. The consultant's job is the part no recording captures: noticing, and writing down what they noticed while it is still fresh. The machine handles what was said. The person handles what it meant.
Interviews recorded, transcribed and summarised into the profile, so the consultant is listening rather than typing.