How AI interviews work: voice, video and the report you get

AI interviews work by asking each candidate structured, role-specific questions built from your job description, recording and transcribing the answers, and scoring them skill by skill against the same rubric. The employer gets a written report for every candidate and decides who moves to the live final interview.

There are two formats. An AI voice interview is a first-round call in English. An AI video interview collects recorded answers and scores them with timestamps. Here is how each one works, what the report contains and how to use it well.

AI voice interview and AI video interview: the difference

An AI voice interview is a first-round voice call. The AI interviewer asks role questions, follows up on what the candidate says and asks the joining questions at the end. It suits high-volume first rounds and roles where spoken communication matters, such as sales, customer support and operations.

An AI video interview asks candidates to record answers to role-specific questions. Each answer is scored, and the report carries timestamps so a reviewer can jump to the moment that matters instead of watching the whole recording. It suits roles where you want to see how someone explains their work, such as technical, client-facing and team lead roles.

Both start from a link sent by email or WhatsApp, and candidates take them on their phone at a time that suits them. No recruiter needs to be online, which removes most of the scheduling back and forth from the first round.

How the questions are built from your job description

The job description is the input. The AI interviewer reads the must-have skills, experience and responsibilities, plans questions that test each skill, and uses the candidate's CV for context. Each question is tied to a skill, so the report can show a score per skill rather than one overall impression.

In a voice interview, follow-up questions adapt to the answers. A candidate who mentions handling escalations is asked how they handled a specific one, and a vague answer gets a probing follow-up, much as an experienced recruiter would ask.

This means the job description sets the quality of the interview. If the must-haves are missing or vague, the questions will be too. The AI Hiring Blueprint can draft a job description from one line about the role, but your team should check and edit it before any links go out.

What the report contains

When a candidate finishes, the recruiter gets a written report on the candidate's card. For an AI voice interview, it covers three things.

Scores: an overall score with a band (strong, good, fair or weak) and a score for each skill, so you can see where a candidate is strong and where they are thin.

Summary and readiness: a short summary of the answers, a job-readiness view and the joining answers, which are notice period, location, expected salary and earliest start date.

Recommendation and evidence: a recommendation to proceed, hold or not proceed, with the full transcript and the recording.

For an AI video interview, the report links each score to a timestamp, so a hiring manager can watch the two or three answers that matter most in a few minutes. Recordings can be shared with your panel, so everyone sees the same evidence before the live round.

Use scores as evidence, not verdicts

A score summarises the answers. It does not know your team, the manager or the trade-offs you are willing to make. Treat it as evidence that points you to the right part of the transcript or recording.

Read the transcript for anyone close to your cut-off before you decide. Compare per-skill scores against your must-haves, not just the overall number. Check the joining answers alongside the skills, because a strong candidate with a notice period you cannot wait for is still a mismatch.

In Million Talents AI, a low score never rejects a candidate by itself. A recommendation to hold or not proceed leaves the candidate at the screened stage for a recruiter to review.

Candidate experience and fairness checks

Candidates should know they are speaking to an AI interviewer. Tell them how the round works, what it covers and roughly how long it takes, and ask for consent before anything is recorded. Letting them choose when to take it respects their time, and that reflects on your employer brand.

Fairness needs checking, not assuming. Every candidate for a role should be assessed on the same skills from the same job description and scored against the same rubric. On top of that, run regular checks.

Review a sample of low scores each week and ask whether a recruiter would have rated them differently. Compare results across groups of candidates for the same role, for example by city, and look closely if one group scores lower without a clear reason. Remove questions that test something the job does not need, such as accent rather than clarity. Keep a person in the loop for every rejection.

Where AI interviews fit before the live final

AI interviews are a first round, not the whole process. A typical sequence is CV screening, a skills test, an AI voice or video interview, then a live final interview where your panel meets the shortlist on Video Meetings. The human makes the final decision.

Use the report to plan that final round. If the summary says a candidate's examples were thin on stakeholder management, probe it live. If the notice period given in the call differs from the CV, confirm it. Your panel's time then goes on the questions that need judgement.

Try it on your next role

Talents AI, the AI hiring assistant from Million Talents AI, runs AI voice and video interviews on your own requirements, alongside skills tests and the live final interview. The AI Voice Interview and AI Final Video Interview pages explain each product, and the Pricing page lists credits per run. Start a free trial to try it: 15 days, ₹5 authorisation today, refunded.