Updated 2026-09-16
An interview engine is software that conducts a structured interview itself — deciding what to ask next, adapting to a candidate's answers, scoring responses, and producing evidence a hiring team can check — rather than just scheduling, recording, or storing an interview someone else conducted.
The term is newer than the products it describes, which is why vendors use overlapping labels — AI recruiter, video interview platform, assessment tool — for partially overlapping capabilities. An interview engine makes a narrower claim: it runs the interview, not just the pipeline around it.
Running an interview in software requires several subsystems working together in real time while someone is speaking. Speech must be captured cleanly. Turn-taking logic must decide whether silence means the candidate is thinking or finished. Speech-to-text must transcribe accurately enough that downstream reasoning is not working from garbage. A reasoning layer must choose follow-up questions based on everything said so far, not a fixed script. Scoring must apply a rubric consistently. Evidence must link each score to citable transcript moments so a hiring manager — or a candidate disputing a result — can verify the claim.
That stack differs materially from an applicant tracking system, which tracks candidates through stages but does not conduct conversations. It also differs from one-way video tools that record answers without adapting, and from sourcing AI that ranks resumes without interviewing.
Capture
Turning microphone input into a clean stream the rest of the system can work with, including detecting when the candidate has started and stopped speaking.
Turn-taking
Continuously deciding whether to keep listening, interject, or wait — the hard case is ambiguous silence after an answer ends.
ASR (speech-to-text)
Converting speech to text accurately across accents and connection quality so the reasoning layer sees what was actually said.
Reasoning
Choosing the next question based on prior answers while holding conversational memory — including noticing contradictions across the session.
Scoring & evidence
Applying a rubric consistently and attaching each judgment to specific transcript moments, not an unexplained number.
Write-back
Exporting score, evidence, and transcript into the ATS or dashboard where recruiters and hiring managers already work.
| Category | Primary job | Conducts live interview? | Adapts in real time? |
|---|---|---|---|
| Interview engine | Runs the structured interview | Yes | Yes |
| ATS | Tracks pipeline stages | No | No |
| Video interview tool | Records fixed prompts | No | No |
| AI recruiter / sourcing | Finds and ranks candidates | No | No |
| Skills assessment | Tests fixed items | No | No |
Procurement teams comparing vendors often receive demos that blur categories — a sourcing product with a chat widget is not an interview engine if it never scores spoken job performance against a rubric. Clarity on what layer you are buying prevents paying for pipeline tracking when you needed conversational screening, or vice versa.
Accent handling and turn-taking quality are not cosmetic features for voice-heavy hiring. Speech recognition tuned primarily on a narrow accent distribution produces bad transcripts; bad transcripts produce irrelevant follow-ups; irrelevant follow-ups feel like the system is ignoring the candidate. Engineering teams building interview engines treat ASR word error rate and voice latency as first-class metrics because they directly affect hiring signal quality.
An interview engine is designed to plug into an existing stack as one stage — feeding structured output into whatever system already tracks candidates, similar to notes from a human phone screen, but at consistent volume and with auditable evidence.
No. An ATS is a system of record for pipeline tracking. An interview engine conducts the interview itself and exports results into the ATS or equivalent workflow tool.
It handles structured, repeatable screening at volume so human interviewers can focus on stages requiring judgment a machine should not make alone — not on removing people from hiring decisions entirely.
The reasoning layer only sees what speech-to-text transcribes. Poor transcription on non-default accents produces bad follow-ups that candidates experience as the system not listening.
A citable link between a score and the specific transcript moment that produced it — so results can be checked against what was actually said.
Products in this category are typically built to export scores, transcripts, and evidence into downstream systems, though integration depth varies by vendor and ATS.
Head back to Hiring glossary or start now.
· 3 free credits · pay per interview · nothing recurring