What is an AI interviewer? A working definition for hiring teams

An AI interviewer is software that conducts a live, two-way interview with a candidate — usually by voice — and scores the answers against a rubric the hiring team defined in advance. It is not a recording tool and not a resume parser: it listens to each answer, asks the follow-up that answer earned, and returns a structured score with the transcript evidence behind it.

That definition is narrower than the market's, deliberately. Plenty of things sold as AI interviewing are really scheduling automation, transcription, or a video recorder with a scoring form bolted on. The line we draw: if the system cannot change its next question based on the answer it just heard, it isn't interviewing.

What makes it an interview rather than a form?

Adaptivity. A form asks everyone the same questions in the same order and stores whatever comes back. An interview responds, which means the same opening question branches three ways depending on the answer.

That branching separates a five-minute answer that reveals nothing from one that reveals a lot. A candidate says they "led the migration." A script moves on. An interviewer picks a probe: depth (what specifically broke?), ownership (which part was yours?), or blast radius (who else had to change what they were doing?) — whichever the answer left unexamined.

It's also what candidates notice. When the follow-up is clearly about the thing they just said, the conversation reads as a conversation — 97% of Lantern candidates finish satisfied with how it went, and what they say about it is mostly about being listened to.

What actually happens during an AI interview?

The candidate joins a voice call, usually from a phone, at a time they picked. The system works through the rubric's criteria conversationally, probing where answers are thin, and ends with a scored, evidence-linked write-up for the hiring team. Step by step:

1. The rubric is locked first. Criteria, levels and evidence standards are agreed before a candidate is invited. Nothing is scored on a criterion that wasn't defined up front.
2. The candidate picks a slot. No coordination step, so the interview happens at 11pm after a shift as readily as at 2pm on a Tuesday.
3. The conversation runs. Question, answer, follow-up drawn from the answer, repeat — in the language the candidate thinks in.
4. Each answer is scored against the rubric, with the score pinned to the transcript moment that earned it.
5. A ranked shortlist reaches the hiring team with the reasoning attached. A human reads it and decides.

How does an AI interviewer score an answer?

It applies the same rubric to every candidate and links each score to the specific thing the candidate said. The value isn't that a machine judges better than a person — it's that it judges the same way at 9am Monday and 6pm Friday, across the hundredth candidate and the first.

Consistency is the honest claim here. Human panels drift: interviewers get tired, calibrate against whoever they saw last, and weight criteria differently without noticing. A rubric applied identically every time removes that variance. What it does not remove is the need for a good rubric, which is a human job and a harder one than it looks — see how to write interview rubrics an AI can score consistently.

Does an AI interviewer decide who gets hired?

Not in our design. Lantern never rejects a candidate. It interviews, scores, ranks, and hands the whole thing to the hiring team, who make every final call.

That boundary matters for more than comfort. It's where the compliance story sits: an automated system that made final selection decisions on its own would be a very different regulatory object. Flags work the same way. When Lantern detects a deepfake, a cloned voice, or an AI-generated script, the flag is never an auto-rejection — the evidence goes to a human with the transcript attached. More in how we approach the EU AI Act.

Where does an AI interviewer fit in the funnel?

It replaces the first conversation, not the last one. The 20-minute screen most orgs run as a recruiter call — the one that pattern-matches against a rubric nobody wrote down — is the slot it's built for.

• High-volume roles where the applicant count makes a human first round impossible without headcount.
• Distributed pipelines where candidates and interviewers sit in time zones that don't overlap politely.
• Multilingual markets where the constraint isn't interviewer capacity but language coverage.
• Any funnel with a scheduling bottleneck, where apply-to-first-conversation is measured in weeks.

It fits badly where the first conversation is a sell rather than a screen: executive hiring, small pools, roles where the pitch is the point. The category isn't universal and we'd rather say so.

What separates a good AI interviewer from a weak one?

1. Does the follow-up depend on the answer? If question order is fixed, it's a recorded questionnaire — see AI interviewer vs. one-way video.
2. Can you see why a score happened? A number without an attached quote is a guess with a decimal point.
3. Who owns the reject decision? If the tool auto-rejects, you've delegated the one act you're accountable for.
4. Does it work in the language your candidates speak? Translating after the fact isn't interviewing natively — see multilingual AI interviews.

Lantern has run 100K+ interviews across 40+ languages and dialects against those four constraints. The numbers we care about are the boring ones: consistent scoring, evidence for every rating, a human on the end of every decision.

Common questions

Is an AI interviewer the same thing as an AI recruiter?

No. "AI recruiter" usually describes sourcing and outreach — finding people and getting them to apply. An AI interviewer starts after the application, handling the conversation and the scoring. Different bottlenecks; a team can have one without the other.

Will candidates take a voice interview seriously?

They already do. Completion tends to go up once the interview stops requiring a weekday slot and a quiet office, because most drop-off is a scheduling problem rather than a motivation problem.

What happens if the system isn't sure about someone?

It escalates. Uncertainty, complaints, accommodation requests and anything the rubric doesn't cleanly cover go to the client's team with the full transcript attached. More answers sit in the landing page FAQ.

Lantern is an AI interviewer built in Hong Kong: adaptive voice interviews in 40+ languages, one rubric for every candidate, a ranked shortlist by morning. To watch one run against a live req, book a walkthrough.