AI Interviews Explained: How They Work, What They Measure, and Candidate Rights

An AI interview is a structured interview conducted by a conversational model. It is not a personality quiz, and the good implementations are not scored on your face or your voice.

How a well-built AI interview runs

1. 01Disclosure. Before anything else, the candidate is told the interviewer is an AI system, that the session is recorded and transcribed, and what happens to that recording. Consent is asked for, and declining is not held against the candidate.
2. 02Calibration to the role. The question set is generated from the scorecard: the four or five competencies that actually predict success in this job.
3. 03Structured questioning. Every candidate gets the same core questions in the same order — the property that makes structured interviews far more predictive than free-form chats.
4. 04Adaptive probing. This is what a form cannot do. A vague answer gets followed up: which part did you build, what broke, what did you change, what would you do differently.
5. 05Evidence linking. Claims are cross-checked against public artifacts and stated history. "Led the migration" is worth more when a commit history agrees.
6. 06Assessment. The output is a written evaluation per competency, each conclusion citing the answer or artifact it rests on, with the gaps named as clearly as the strengths.
7. 07Human review. A person reads it, weighs it, and decides. The system ranks; it does not hire.

What it measures

Reasoning under a follow-up. Depth of ownership of past work. Specificity — the difference between someone who was in the room and someone who read the postmortem. Communication clarity. Role-specific technical judgement.

What it should never measure

Facial expression, perceived emotion, accent, or speech rate. These have weak-to-no evidence for job performance and carry serious discrimination risk. If a vendor sells "emotion AI" for hiring, that is a reason to walk.

Candidate rights, practically

• Notice that AI is being used, before the interview.
• An accommodation path — a human interview or adjusted format on request, without penalty.
• Explanation of what was assessed and, where a decision is challenged, the evidence behind it.
• Human review of an adverse outcome.
• Data control — knowing retention periods and being able to request deletion.

Regulation is catching up: the EU AI Act treats recruitment systems as high-risk with transparency and oversight duties, and NYC Local Law 144 requires a bias audit and candidate notice for automated employment decision tools. Treat these as the floor.

How to prepare as a candidate

Prepare specifics, not stories. Have three pieces of work you can describe at the level of decisions you personally made. Expect follow-ups and welcome them — the probing is where you gain ground over a stronger CV. Answer the question that was asked; the model is patient but literal. Say "I don't know" when you don't; a clean admission scores better than a bluff that unravels over two follow-ups.

For employers

Publish the competencies you assess. Keep the interview under 30 minutes. Give every candidate their assessment summary — it is the cheapest employer-brand investment available and it is what candidates say they miss most from traditional processes. And keep a human on every decision.

Octively runs interviews on exactly these rules: disclosed, structured, evidence-cited, human-reviewed. Join the waitlist below.