AI interviews: how they work and how to use them properly

An AI interview is a structured interview a candidate completes on their own, usually by recording answers to set questions, where an AI system transcribes the responses and scores them against defined criteria. It sits where a phone screen sits: after the application, before a human interview.

The category has a poor reputation, largely earned. Early products claimed to infer personality and suitability from facial expression and vocal tone, which was not well supported and produced predictable discrimination problems. Understanding what the technology legitimately does, and refusing the part it does not, is most of what an employer needs here.

What it actually assesses, and what it must not

Legitimate: what the candidate said. The content of the answer, assessed against the criteria you defined. Whether they addressed the question, whether the example was relevant, whether they can explain their own reasoning, whether they meet stated requirements. This is a structured interview with the transcription and first-pass scoring automated.

Not legitimate: inferring capability, personality or "fit" from face, voice, tone or body language. The evidence base for this is weak, and the discrimination risk is severe and direct, because it penalises accent, neurodivergence, disability, and anyone whose presentation differs from the training data.

HRM8's line on this is explicit. Face and voice are used for identity verification and anti-cheating only, with explicit consent and short retention. They are never used to score suitability or infer emotion. If a vendor markets emotion detection or personality inference from video as a selection feature, treat that as disqualifying.

Where it earns its place

Volume. Sixty applicants and a screening call each is roughly twenty hours. An AI interview gives every one of those sixty the same questions and gives you a consistent basis to compare, at a cost per candidate that makes it viable.

Consistency. Phone screens are the least structured stage in most processes. The first five and the last five get different interviews from a tired person. Automating that stage removes an inconsistency people rarely think about.

Scheduling. The single largest source of elapsed time in most processes. Asynchronous completion removes the coordination problem entirely, and candidates in other time zones or in a current job can complete it when it suits them.

Reach. Candidates who cannot take a call during working hours are systematically disadvantaged by live phone screens. Asynchronous helps them.

Where it does not belong

As the only interview. It is a first round. Nobody should be hired without a human conversation.

As an automatic rejection mechanism. This is the important one. The system narrows and orders, and a person decides. HRM8's Hub never auto-rejects, and any disqualifying result routes to the employer to sign off.

For senior or specialist roles. Where the field is small and the assessment is nuanced, the efficiency argument disappears and the candidate-experience cost rises. A senior candidate asked to record answers into a camera for a first round will often simply decline, and they will be right to.

Where the role requires live interaction. If the job is live client conversation, assess live conversation.

The obligations that now attach

This is a regulated area and the requirements are becoming concrete.

The EU AI Act classifies AI used in recruitment and selection as high risk, with obligations covering risk management, data governance, record-keeping, transparency and human oversight. New York City's Local Law 144 requires an annual independent bias audit of automated employment decision tools, public disclosure of the results, and notice to candidates in advance. Colorado's AI legislation covers consequential decisions including employment. Australian privacy law covers automated decision-making, and UK regulators have issued guidance on AI in recruitment.

Four things to hold regardless of where you hire:

1. Tell candidates before they start: that AI is used, what it assesses, and that a human makes the decision. 2. Offer an alternative for candidates who cannot complete a video assessment, and make requesting it easy and private. 3. Keep the human decision real, not nominal. A human who rubber-stamps a ranked list is not meaningful oversight. 4. Keep records: questions asked, criteria, scores, the decision and who made it.

Ask any vendor how scores are produced, whether the tool has been bias audited and when, what data is retained and for how long, and whether the system can reject anyone without a human. The last answer should be no.

Candidate experience is the real risk

Recorded interviews are widely disliked. Candidates describe them as impersonal and one-sided, and strong candidates with options drop out.

What reduces the damage:

Keep it short, four to six questions. Explain why you use it and what happens next. Show the questions before recording starts. Allow a retake. Give a genuine deadline rather than an artificial one. And respond to everyone, because an unanswered recorded interview is worse than an unanswered application.

The clearest signal of good intent is a human contact point. A candidate who can email a person about the process experiences it very differently from one shouting into a form.

What it costs

HRM8 prices an AI interview at US$25 per candidate, or US$145 for a job pack of ten, which is US$14.50 each. There is no subscription, and credits do not expire.

Against a screening call at twenty minutes of a manager's time plus the scheduling overhead, the arithmetic is usually favourable at volume and usually not worth it below about fifteen candidates, where a human call is better anyway.

AI interviews are also built into HRM8's Recruitment Services engagements, where a recruiter uses them to inform screening and the recruiter and employer own the shortlist decision.

Common questions