An async AI interview is a recorded interview conducted by AI instead of a scheduled call: the candidate opens a link whenever they're ready, the AI asks questions aloud and adapts to the answers, and every response is transcribed and scored against the role's rubric. No calendars, no panels, no timezone algebra — and, done right, more consistent evaluation than the calls it replaces.
This piece walks the whole mechanism: what the candidate experiences, what the AI actually does, what the recruiter receives, and where async genuinely beats scheduled screens (plus where it doesn't). For where the async interview sits in the wider screening stack, start with what an AI ATS is.
Why the first interview was always the bottleneck
The first-round screen has brutal arithmetic. Thirty minutes of interview costs about an hour of recruiter time with scheduling, notes, and context-switching. Two hundred applicants means you screen twenty and guess about the rest — which means the real first-round filter at most companies isn't an interview at all; it's whatever CV skim decided who got the twenty slots. Async AI interviews attack the arithmetic directly: interviews run in parallel, on candidates' own time, conducted and scored by the machine. Screening a hundred applicants costs the same recruiter time as screening five — near zero — and the recruiter's hours move to reviewing ranked, evidenced results.
The candidate's side, step by step
In Rubrily the flow looks like this: the candidate receives a branded invitation showing the role, format, and duration; picks an interview language from the options the employer enabled; can take a free practice walkthrough before anything is graded; answers pre-screening questions and uploads a CV; gives explicit consent to recording after seeing the monitored-interview conditions; and then takes the interview itself — questions narrated aloud, answers spoken, a visible timer, progress saved as they go. The AI may ask a follow-up based on what the candidate actually said.
Details in that list carry more weight than they look. Language choice means candidates compete on substance, not on English confidence. Practice-before-grading means the tool doesn't penalize unfamiliarity with itself. Consent-before-recording is both the legal floor and the trust signal. We've written the whole experience from the candidate's point of view — worth reading before you ship your first invite, because your process is their first impression of you.
What the AI is doing
Three things distinguish an AI interviewer from a survey form with a webcam:
It conducts, not just collects. Questions come from an interview set the recruiter defines once — with per-question priority, so the AI spends time where the role needs it. It adapts. Based on the candidate's answer, the AI can cross-question — probe the vague claim, ask for the concrete example — the way a good human screener would, while staying inside the same rubric for everyone. It scores with reasons. Every answer is transcribed and scored 0–10 per criterion with a written justification citing what the candidate said. Those scores roll up through the rubric's weight tiers into an interview score, which blends with the CV evaluation into the candidate's Fit Score.
Integrity runs underneath: tab switches, extended absence from camera, and AI-generated answers can be detected and surfaced to the recruiter as flags on the report — signals for a human to weigh, not tripwires that auto-reject.
Do candidates need to schedule anything?
No — that's the "async" in async AI interviews. The invite link works whenever the candidate is ready: evening, weekend, any timezone. Interviews run in parallel across the whole applicant pool instead of serially through a recruiter's calendar, which is how days of screening compress into hours of review.
What the recruiter receives
Each completed interview lands as a scored submission: overall 0–10 with a narrative summary citing evidence, the per-criterion breakdown with justifications, strengths and gaps, the full timestamped transcript, and the recording itself. Candidates rank in the pipeline by Fit Score, and the whole report is shareable with your team — comments, ratings, decision. You're not trusting the AI's judgment; you're reading its work.
Where async wins — and where it doesn't
Async is the right tool for the first interview: high volume, comparable answers, evidence for who deserves human time. It is not a replacement for the conversations that follow — the deep technical dive, the team-fit discussion, the sell call. Those need humans, and the honest pitch for async AI interviews is that they fund those conversations: every hour not spent on screening calls is an hour available for the interviews that actually decide.
Teams sometimes worry that async screening will feel cold to candidates. The comparison to make is with the alternative at volume — which for most applicants is silence followed by a form rejection, no interview at all. An async interview means every applicant gets heard, in their own language, at their own time, and evaluated on the same criteria as everyone else. That's not colder than the status quo. It's the first time most of the pile gets warmth at all.
FAQ
How long should an async AI interview be? Twenty to thirty minutes covers a 5–8 criterion rubric with room for follow-ups. Shorter starves the rubric of evidence; longer taxes completion rates. Set the duration per interview set and show it up front — candidates plan around what they can see.
Can the AI really ask follow-up questions? Yes. Adaptive cross-questioning reacts to the candidate's actual answer — probing deeper on what they said rather than reading a fixed script — and can draw on the candidate's CV. Interviews stay comparable because the rubric and criteria stay fixed; the path through them flexes.
What if a candidate's connection drops mid-interview? Progress saves as they answer; reopening the link continues where they left off. Incomplete interviews are scored on what exists or marked "Cannot evaluate" — a dropped connection is treated as missing data, never as a failed answer.
What stops candidates from cheating? The session is monitored with the candidate's informed consent: tab switching, leaving the camera, and AI-generated responses can be flagged. Flags surface to the recruiter alongside the recording — a human weighs them in context, and nothing auto-rejects.
Set your questions once — the AI interviews every applicant, async, in their language, and scores every answer with reasons. Start free →