Using AI to Analyse Interviews Fairly: Consent, Bias and Regulation Basics

Using AI to analyse interviews can be done fairly if the candidate knew the interview was being transcribed, the AI is restricted to job-related evidence, a person makes the decision, and you have checked the rules that apply where you hire. This article is general information, not legal advice; confirm your obligations with qualified counsel.

1. Consent and notice

A transcript contains a candidate's personal data. Tell candidates in advance that the interview will be recorded or transcribed and what it is used for. Recording laws differ by place, and some require everyone on the call to agree. Only analyse transcripts of interviews where that was the case.

2. Keep the AI to job-related evidence

The main bias risk is a model commenting on things unrelated to the job. Instruct it explicitly to judge only skills, experience and reasoning, and never to infer or mention characteristics such as age, gender, ethnicity, accent, health or family status. Output that refers to the candidate neutrally avoids a common slip: assuming gender from a name.

3. Require evidence for every claim

A tool that outputs a score with no justification invites over-trust. Prefer evaluations that tie each judgement to something the candidate said, and verify that quoted text actually appears in the transcript. An unsupported quote should be removed, not shown.

4. Keep a human in the decision

Treat AI output as decision support. A recruiter should be able to read the evidence, disagree, and change the rating, and nothing should reject or advance a candidate automatically. Human review is also what most regulations around automated hiring tools emphasise.

5. Check the regulations that apply to you

Rules vary and are changing. Examples worth reading are New York City's Local Law 144 on automated employment decision tools, which involves bias audits and candidate notice; the EU AI Act, which treats AI used in recruitment as high-risk; and Illinois' law on AI analysis of video interviews. Whether and how each applies depends on how a tool is used, so get advice for your situation.

6. Limit retention and allow deletion

Keep transcripts only as long as needed, restrict who can see them, and delete them, along with any evaluation, on request or when your retention period ends.

How Rezumz approaches it

Rezumz's Interview Transcript Analysis is designed around these points: a consent reminder before upload, evaluation limited to job-related evidence, no gendered pronouns, verified quotes, a suggested rating the recruiter edits, no automatic status changes, and deletion at any time. Using it does not remove your own legal obligations.