An AI video interview is a one-way, asynchronous video interview in which candidates record answers to set questions and AI transcribes and scores those recorded answers to help recruiters rank applicants. It is the AI-powered version of a standard one-way interview: the recording step is the same, but software reads the transcript and rates each answer against defined criteria before a person reviews the shortlist. This guide explains how the process works, how it compares to a plain one-way interview, and what AI should and should not be trusted to judge.
Key takeaways
- An AI video interview is the AI-powered form of a one-way interview: candidates record answers to set questions, and AI transcribes and scores them.
- AI is good at transcribing, summarizing, matching answer content to a rubric, and ranking candidates so review takes less time.
- AI should score what a candidate says (content, structure, skill match), not how they look or sound: facial and emotion analysis is discredited and restricted in some places.
- Keep a human in the loop. Treat AI scores as a first pass to review and audit, not a hiring decision.
How AI video interviews work
Start with the one-way interview. In a one-way (or asynchronous) interview, there is no live interviewer. The candidate reads a question, records a video answer on their own time, and submits it. The recruiter watches the recordings later.
An AI video interview adds a layer on top of that recording. After the candidate submits, AI converts speech to text and evaluates the answer content against criteria the hiring team set. The output is a transcript, a summary, and a score or ranking for each candidate.
How an AI video interview works, step by step
- 1
The candidate records answers to set questions
Every applicant gets the same questions and records a short video response to each, usually with a time limit and a set number of retakes.
- 2
AI transcribes the recordings
Speech-to-text turns each spoken answer into a written transcript the hiring team can read and search.
- 3
AI scores the content against criteria
The system rates each answer against a rubric the team defined, for example whether the answer covers required skills, uses a clear structure, or names relevant experience.
- 4
The recruiter reviews the ranked shortlist
The recruiter sees candidates ordered by score, reads transcripts and summaries, watches the strongest recordings, and decides who moves forward.
The value is speed at the top of the funnel. Instead of watching 200 raw recordings, a recruiter reviews a ranked list with transcripts and can focus attention where it matters.
How it differs from a plain one-way interview
A plain one-way interview and an AI video interview use the same recording format. The difference is entirely in what happens after the candidate submits.
| Capability | Plain one-way interview | AI video interview |
|---|---|---|
| Candidate records answers | Yes | Yes |
| Transcript of each answer | Manual, if done at all | Automatic |
| Answer summary | Recruiter writes notes | Generated for each answer |
| Scoring against a rubric | Recruiter scores by hand | AI scores first, recruiter reviews |
| Candidate ranking | Manual comparison | Ranked list out of the box |
| Review time per candidate | Full recording watched | Skim transcript, watch selectively |
In both cases a human still makes the hiring call. The AI version changes how quickly you can get to that call, not who makes it.
What AI should and should not evaluate
The line to hold is simple: score what a candidate says, not how they look or sound. Content maps to job performance. Appearance and delivery do not, and judging them invites bias and legal risk.
- Good use: the substance of the answer, whether it addresses the question, and whether it shows the required skills.
- Good use: structure and clarity, for example whether a problem-solving answer walks through a logical approach.
- Good use: keyword and skill match against the role, treated as a signal to review rather than a verdict.
- Avoid: facial expression and emotion analysis, which is scientifically discredited and restricted by law in some jurisdictions.
- Avoid: generic employability or personality scores that are not tied to specific, job-related criteria.
- Avoid: judging accent, speech pace, or tone of voice, which penalizes non-native speakers and people with disabilities.
If a vendor sells emotion detection or a single employability score, ask exactly what it measures and how it was validated. Vague answers are a reason to walk away.
Fairness and keeping humans in control
Practical guardrails: tell candidates AI is used and offer an alternative if they request one, keep questions job-related and consistent, review the rubric with more than one person, and check periodically whether scores differ by gender, ethnicity, or age in ways the job cannot explain.
Used this way, an AI video interview saves review time and adds structure to early screening. Used as an unchecked filter, it scales bias. The difference is the human who stays in the loop.
What the AI is actually reading
The transcript, not the face. That distinction matters more than any other in this category. A responsible AI video interview transcribes what the candidate said and scores the content: whether it answers the question, whether it contains specifics, how clearly it is expressed, and how well it matches the role. It does not infer personality from facial expressions, does not score enthusiasm from tone, and does not penalise an accent. Systems that claimed to read emotion from faces are why this category earned its regulatory scrutiny, and the science behind those claims did not hold.
- Scored: the substance of the answer, its relevance, its specificity, its clarity.
- Not scored: appearance, background, accent, facial expression, perceived confidence.
- Never automatic: rejection. Only a knock-out question about a fact can remove a candidate without a person deciding.
Questions to ask a vendor before you buy
- Does every score link to the transcript that produced it, so I can check the working on any candidate?
- What exactly is scored, and is anything inferred from the video rather than the words?
- Can the system reject a candidate without a human, and can that be switched off?
- Am I charged per invitation or per completed interview?
- What happens when a recording fails: is a silent answer detected, or scored as a bad one?
- Can a candidate ask for a human review, and what does that process look like?
