HireVue AI: How the Algorithm Actually Scores Your Interview
HireVue's AI stopped scoring faces in 2021. Here is what the algorithm actually evaluates today: transcribed answer content against a per-role competency rubric. Plus what scores top band.
HireVue's AI does not score your facial expressions. It did, until 2021. After three years of academic critique, an FTC complaint, and the company's own internal validation showing facial analysis added nothing to predictive accuracy, HireVue removed all visual cues from the model. What the algorithm scores today is the transcript of what you said, mapped against a competency rubric the employer built for the role.
This article walks through what the current HireVue AI actually evaluates, how the transcription-to-score pipeline works, what the recruiter sees on the other side, and what candidates can do to score in the top band given a system that no longer cares about eye contact or smiling.
Quick takeaways
- HireVue's AI scores the transcribed text of your answers, not your face, voice tone, or body language. Facial analysis was removed in 2021.
- A natural-language model maps your transcript against a competency rubric the hiring team set up for the specific role. Most roles measure 5 to 7 competencies.
- Candidates are placed into Top, Middle, or Bottom bands relative to other applicants for the same role. The dashboard the recruiter sees is per-competency, not a single number.
- The same competency rubric is applied to every candidate, which is the basis for HireVue's "removes interviewer bias" pitch.
- The single biggest scoring lever is on-topic specificity. Generic answers score in the middle band even when delivered confidently. Specific, situation-anchored answers (the STAR method) score in the top band.
- Recording quality matters because it affects transcription quality. Poor audio leads to garbled transcripts and incomplete scoring.
- The AI does not "decide" whether you advance. It produces a ranked shortlist that a human recruiter then reviews. About 60 to 80 percent of HireVue-scored candidates still receive a human screen before any hire decision.
What HireVue used to measure (and why it stopped)
Until January 2021, HireVue's video assessment scored four things: speech content (transcribed words), vocal cues (pitch, pace, energy), facial cues (smile rate, eye contact, micro-expressions), and language patterns. The platform marketed this as "20,000 data points per 30-minute interview." The pitch was that machine vision and speech analysis could pick up subtle signals human interviewers miss.
The pitch ran into three problems. Academic researchers showed that facial-cue scoring discriminated against candidates with darker skin (poorer landmark detection), candidates wearing glasses, candidates with neurodivergent expressions, and candidates with regional accents on the audio model. The Electronic Privacy Information Center filed an FTC complaint in 2019. And HireVue's own internal audit, conducted by O'Neil Risk Consulting, found that facial-cue scoring added almost nothing to the predictive validity already captured by the language model.
In January 2021, HireVue announced it had removed visual analysis from the model entirely. The current scoring engine ingests audio, transcribes it to text using automatic speech recognition, and scores the text. The video portion of the recording still exists, because recruiters watch it, but it does not feed the AI score.
What the AI scores in 2026
The 2026 model is a natural-language pipeline that takes the transcript of your audio response and produces a per-competency score. The competencies are not generic. They are defined by the employer when the assessment is built for a specific role.
A typical role rubric might measure: Customer Orientation, Problem-Solving, Teamwork, Resilience, Communication, Initiative, and Adaptability. The hiring team selects competencies from a HireVue library and the platform constructs interview questions designed to elicit answers that demonstrate each one.
Per competency, the model has been trained on transcripts from previous high performers and low performers in the same role at companies with enough hiring volume to produce a training set. The model produces a score from 1 to 5 on each competency, then aggregates them with the employer-set weighting.
The diagram below shows where the AI now stops and starts.
The model does not look at filler words ("um", "uh"), grammar, accent, or whether you said something clever. It looks at whether your answer contains the kinds of content patterns that high performers in this specific role tend to produce when asked this specific question. Answers about "leading a difficult project" score higher when they describe a specific project with a specific obstacle and a specific resolution than when they describe leadership in the abstract, because that is what the training transcripts show.
How the transcription pipeline affects your score
Before the language model sees anything, automatic speech recognition (ASR) converts your audio to text. If ASR fails or produces a garbled transcript, the language model scores garbage. There are four ways candidates accidentally hurt their transcription quality:
The first is poor microphone. Built-in laptop microphones with the laptop fan running produce noisy audio. Modern ASR is robust, but candidates who use a clip-on or USB microphone get cleaner transcripts than those using a laptop in a coffee shop.
The second is over-fast speech. ASR systems get markedly worse above 180 words per minute. Anxious candidates speed up. The transcript drops words. The score drops with it. A measured 150 words per minute reads cleanly.
The third is heavy regional accent on a poorly trained ASR model. HireVue uses multi-regional ASR but Latin-American Spanish-accented English or thick Indian-English accents can still produce 5 to 10 percent more transcription errors. Speaking slightly slower and enunciating word endings compensates.
The fourth is talking off-mic. Candidates who turn their head while gesturing drop the audio level. Keep the head pointed at the camera and microphone the entire time.
The competency rubric, decoded
A HireVue assessment for a customer-facing role might ask: "Tell me about a time you turned a frustrated customer into a satisfied one." That question is mapped to two competencies: Customer Orientation and Problem-Solving.
A top-band answer for Customer Orientation contains: specific customer description, a specific frustration (with a number or duration), a specific listening or empathy action, and a specific outcome that recovered the relationship. A top-band answer for Problem-Solving contains: a clear problem statement, a structured approach, an action with constraints, and a measurable result.
A middle-band answer contains general claims ("I am very customer-focused") without specifics, or specifics without resolution.
A bottom-band answer is off-topic ("Customer service is important because..."), too short to score, or fundamentally non-responsive.
| Competency | Top band signal | Middle band signal | Bottom band signal |
|---|---|---|---|
| Customer Orientation | Specific customer, specific issue, specific resolution, recovered outcome | Generic empathy claims, no specifics | Off-topic, defines customer service abstractly |
| Problem-Solving | Stated problem, structured approach, action under constraints, measurable result | Steps without constraints, vague result | No structure, just narrative |
| Teamwork | Names the team, your specific contribution, conflict handling, joint outcome | "I worked with my team to..." with no specifics | Talks about yourself only, no team mention |
| Resilience | Specific setback, emotional acknowledgment, recovery action, lesson applied later | "I bounced back" without context | "I never fail" or denial of setback |
| Communication | Audience-adapted message, clarity check, outcome confirming the message landed | Repeats the message, no audience adaptation | Off-topic or rambling, no clear audience |
| Initiative | Unprompted action, identified need, resource navigation, outcome | "I always go above and beyond" without example | Asked for permission, waited for instruction |
| Adaptability | Specific change, emotional response named, adaptation behaviour, learned skill | "I am flexible" without instance | Resisted the change |
The rubric is the same one a human interviewer would informally use. The difference is that the AI applies it consistently across every candidate in the same role.
What the recruiter sees
After your interview is scored, the recruiter opens a dashboard that shows a band placement (Top / Middle / Bottom) and a per-competency 1-to-5 score. They also see the video, the transcript, and the question-by-question breakdown.
The dashboard view below shows what a recruiter sees per competency.
Most recruiters do not watch every video. The standard workflow is to review the top-band candidates in full and to spot-check a sample of the middle band. The bottom band is usually deprioritised or rejected, depending on volume. About 60 to 80 percent of HireVue-scored candidates still get a human screen before any final decision, but landing in the top band roughly doubles the probability that a human will actually watch your video.
How to score in the top band
Three practical reflexes raise the average HireVue score by enough to move from the middle to the top band for most candidates.
The first is the STAR pattern, applied tighter than the general advice. Situation in 1 sentence, Task in 1 sentence, Action in 2 to 3 sentences with specifics (number, duration, constraint), Result in 1 to 2 sentences with a measurable outcome. The full answer fits in 90 seconds, which is the length most HireVue prompts allow. Long, meandering answers score below tight ones because they bury the high-signal content.
The second is naming the competency without saying its name. If the question is about resilience, do not say "this shows my resilience." Instead, describe a setback explicitly, the emotion you felt, the recovery action, and the lesson. The competency name is for the rubric, not the candidate.
The third is reading the question for the implicit competency. "Tell me about a difficult colleague" maps to Teamwork and Communication. "Tell me about a time you missed a deadline" maps to Resilience and Initiative. Two minutes of role research before the interview tells you which 5 to 7 competencies are on the rubric, and which competency each question is likely targeting.
Practice patterns that build the right reflexes
A 5-day prep plan moves most candidates from middle to top band.
Day 1: Identify the 5 to 7 likely competencies for the role from the job posting and competency frameworks for similar roles. Write a one-line definition of each.
Day 2: Write one STAR-format story per competency. Two paragraphs each. Real situations, real numbers, real outcomes.
Day 3: Record yourself answering each of the 7 likely questions in 90 seconds. Play it back. Check whether the STAR shape is audible.
Day 4: Re-record. Tighten the Situation and Task to one sentence each. Stretch the Action with one more concrete detail.
Day 5: One light practice round in HireVue's official candidate experience. Sleep. Take the real assessment with reflexes installed.
The candidates who go from middle to top band are not learning new content. They are tightening the structure so the AI rubric can detect the signal.
Related on PrepClubs
- Pillar. HireVue overview. The full breakdown of HireVue's assessment, interview, and games products.
- Deep practice. HireVue practice pack with Pass Guarantee. Full mock interviews with scoring breakdown. $39 one time. Pass Guarantee.
- Interview format. HireVue Interview: Format, Questions, and How to Pass. The full walkthrough of the on-demand video interview.
- Brand pillar. HireVue in 2026: Assessment, Interview, Games, and Companies. What all four HireVue products do.
- Companies. Companies That Use HireVue. Industry breakdown of HireVue's employer base.
Practice on PrepClubs
Practice the STAR pattern against mock HireVue prompts with feedback.
The AI scores transcript content against a competency rubric, so the highest-leverage practice is delivering tight STAR-structured answers to prompts the platform actually uses. PrepClubs runs HireVue-style mock interviews with prompts mapped to the same 7-competency framework, plus playback with structured feedback on Situation, Task, Action, and Result clarity. $39 one time. Pass Guarantee.
FAQ
Does HireVue still analyze my face?
No. Facial analysis was removed from the scoring model in January 2021 after internal validation showed it added no predictive accuracy and external research showed it introduced bias against several candidate groups. The recruiter can still watch your video, but the AI does not score it.
What does the HireVue AI actually score?
The transcribed text of your audio answers, mapped against a competency rubric defined by the employer for the specific role. The output is a per-competency score from 1 to 5, plus a band placement of Top, Middle, or Bottom relative to other candidates for the same role.
Does word choice or grammar affect my score?
Not directly. Filler words, mild grammar errors, and accent do not penalize the score. What the model looks for is content patterns: did you describe a specific situation, did you describe a specific action, did you describe a measurable outcome. Tightly structured answers score higher than fluent but vague ones.
Will a human recruiter still review my interview?
In most cases, yes. About 60 to 80 percent of HireVue-scored candidates receive a human screen before any final decision. Landing in the top band roughly doubles the probability that a human will actually watch your full video rather than just glancing at the score.
Can I retake a HireVue interview?
Retake policy is set by the employer, not by HireVue. Most employers allow only the first attempt. A small minority allow a second attempt if the recording had a technical fault. Treat the first sitting as your only one.
Does talking faster or slower help?
Slower helps. ASR transcription quality drops above 180 words per minute. A measured 140 to 160 wpm produces the cleanest transcript and gives you time to actually deliver the STAR structure inside the 90-second window most HireVue prompts allow.
What is the single highest-leverage prep activity?
Writing one STAR-format story per likely competency, then recording yourself delivering each in under 90 seconds. The mechanical act of speaking the story aloud, listening back, and tightening it for the second take installs the structural reflex the AI rubric is built to detect.
Is HireVue legally allowed to use AI to score me?
Yes, with caveats. Illinois (Artificial Intelligence Video Interview Act, in force since 2020) and Maryland require explicit candidate consent and disclosure that AI is being used. New York City Local Law 144 requires bias audits of any automated employment decision tool. HireVue complies in all three jurisdictions, and the consent screen at the start of the interview is the formal mechanism.
FAQ
