AI interviewer, explained for recruiters

If you Google "AI interviewer" you get two kinds of pages: marketing pages selling one, and op-eds about whether you should be allowed to. Neither tells you what an AI interviewer actually does in your week.

This one does. No hype. Just: what it is, what it asks, what you get back, and what it deliberately leaves to you.

What an "AI interviewer" actually is

Stripped of marketing language: an AI interviewer is a piece of software that calls a candidate, has a real conversation with them for a set amount of time, transcribes everything that was said, and scores the conversation against a rubric you wrote.

That's it. Three jobs:

1. Conduct the call — voice, real-time, two-way. The candidate talks; the AI listens, asks the next question, follows up on weak or vague answers.
2. Transcribe and record — every word, both sides, kept for you and the candidate.
3. Evaluate — score each rubric criterion you defined, with the transcript quote that earned the score attached.

What you don't get from a category-honest AI interviewer:

• A "hire / no-hire" verdict. That's your call.
• A keyword-matched resume score with a chat interface on top.
• A one-way video upload the candidate stares into.
• A personality inference from tone, face, or background.

Anything that does the last three is solving a different problem.

What the call sounds like

Our AI interviewer (sample transcript here) runs a 12-minute voice conversation. The first 60 seconds are housekeeping: it identifies itself as an AI, confirms the candidate consents to being recorded, explains that they'll receive their own transcript at the end, and previews what the rubric covers. Then it asks the first question.

A real excerpt from the public sample report (demonstration data — fictional candidate, real production format):

AI: You mentioned Kafka earlier — how much of that pipeline was Kafka in production?
Candidate: …Honestly, mostly NATS. I did one Kafka migration at my last job, but day-to-day it was NATS.

The follow-up matters more than the original question. The candidate said "Kafka" earlier; the interviewer noticed; it came back. That's the difference between a voice agent that listens and a chatbot that runs a script. A resume parser scores the word "Kafka." A real conversation scores what's behind it — and writes down the correction.

What lands in your inbox

For every candidate, you get one report. From the same sample:

7.8 / 10 — Advance to technical round
Weighted rubric score across 4 criteria. Recommendation threshold for this campaign: ≥ 7.0 advance, 5.0–6.9 review, < 5.0 decline.

Two design choices in that header matter to your work:

• The score is weighted against your rubric — the criteria and weights you set on campaign creation, not a generic "communication skills" template baked into the vendor.
• The advance threshold is yours. If your bar for "advance" is 8.0, set it to 8.0. The job of the AI interviewer is to apply your judgment consistently across every candidate, not to substitute its own.

Below the header, every criterion has a score, a weight, and the transcript line that earned it. From the same sample:

Calibration — weight 2 — 6/10
Evidence (from transcript): Pushed back well once (scope of on-call ownership) but overclaimed on Kafka experience; when the agent followed up, admitted it was "mostly NATS, one Kafka migration." Honest on the follow-up, but the initial claim needed the push.

The 6/10 is the part that should tell you the product works. The candidate overclaimed; the report doesn't round it up to make the AI look smart. When you disagree with a score, the quote is right there to argue with.

Below the scores: a short list of strengths and flags for the human round ("probe Kafka/NATS depth directly; candidate hedged on pager duty"). That's the handoff most screening processes lose — the recruiter who did the phone screen knows the candidate hedged on pager duty, but the hiring manager doing round two never hears about it. Here it's written down, attached to the candidate, before anyone's memory fades.

What the candidate gets

This is the part most "AI interviewer" pages don't talk about, because most candidates don't get anything.

Ours do. The candidate keeps their full transcript and sees their own scored report — the same one you see. If you're going to score someone with software, they should be able to read the score. It also keeps the rubric honest: every number you give is a number the candidate can read.

That's not a feature, it's a position. If you're shopping AI interviewers, ask every vendor what the candidate walks away with. The answer tells you most of what you need to know about how the product treats your employer brand.

What an AI interviewer does not replace

A useful AI interviewer is one that's honest about its boundaries.

• It does not make hiring decisions. New York's Local Law 144 and the EU AI Act are pushing the whole category in this direction anyway: humans decide; the software supports. The AI interviewer recommends advance / review / decline against your thresholds for the first round. The hire stays with you.
• It does not handle every interview round. First-rounds — the ones eating your week — yes. Cultural deep-dives, executive panels, take-home reviews — no. Anyone selling you "AI for every round" is solving a different problem.
• It does not source candidates. It interviews the ones you have. Sourcing is a separate product category; treat it that way.
• It does not infer character from how someone looks or sounds. No tone analysis, no facial inference, no background scrutiny. The score is built from what the candidate said, quoted back to you. Anything beyond that is a different (and noisier) product.

If your existing process already does some of those things, an AI interviewer doesn't replace them. It replaces the part of your week that is the same first-round, repeated 40 times.

How an AI interviewer fits into your week

Concretely, for the recruiter side, the flow is four steps:

1. Create the campaign. Job title, JD, evaluation rubric (criteria and weights), advance threshold. Five minutes for a role you already know.
2. Add candidates. CSV, paste, or manual. Each candidate gets an invite with the AI disclosure and consent notice.
3. Let the interviews run. The candidate clicks the link when they're ready; the AI runs the call; the report lands when it ends.
4. Read the report. Advance the strong ones, review the borderline ones, decline the rest — and the flags go to the next interviewer with the candidate.

The work you save is the 40 nearly identical first-round phone screens. The work you keep is the part where judgment matters.

How to compare AI interviewers (ours included)

If you're evaluating tools in this category, the report artifact is the fastest way to cut through demos. Ask every vendor — including us — these four questions:

1. Can I see a full sample report before I sign up? Not a screenshot — the actual artifact, scores, evidence, transcript. (Ours is here.)
2. Are scores tied to my rubric or a generic one? If you can't set criteria, weights, and thresholds per role, you're buying someone else's opinion of your candidates.
3. Does every score come with evidence from the transcript? A number without a quote is unauditable.
4. What does the candidate get? A first round sets the tone for your whole process.

We keep honest comparison pages against the other tools in the category — including what they do better than us and when you should buy them instead. Most vendor comparison pages are sales documents. Ours include the parts that aren't flattering.

Try the artifact, not the pitch

The fastest way to evaluate an AI interviewer is the way you'd evaluate a candidate: evidence first.

1. Read the sample scored report — two minutes.
2. Run your first interviews on a real role. First 3 interviews are free, no card, no subscription.
3. Compare the report against your last manual phone-screen notes. Keep whichever is better.