How HireVue Scoring Works: What the AI Reads, and What It Ignores

HireVue scores the words you say, not your face. Your recorded answer is transcribed, and the transcript is scored against a competency model the employer configured for the role: does the content address the question, is it structured, is it specific, and how clean is the delivery in terms of pace and filler words. The score produces a ranked shortlist that human reviewers look at alongside the recording. HireVue does not make the hiring decision; it decides who a person watches first.

• Competency mapping. Each question is built to probe a competency (communication, problem-solving, conscientiousness, teamwork, or a role-specific one). The model reads your transcript for evidence of that competency. A leadership question answered with a story that never names a decision you made scores low on leadership regardless of how it sounded.
• Content relevance. Semantic overlap between your answer and the question's intent. A charming answer that misses the point scores below a plain answer that hits it.
• Structure and specificity. Answers with a clear situation, action, and result, and with concrete details, score above generalities. This is why the STAR method carries more weight here than in a live interview, where an interviewer can prompt for the missing piece.
• Speech signals. Pace, filler-word density, sentences that trail off. These feed a clarity component. They are measured from the audio, and they are the part most candidates can change fastest with practice.

Facial expressions, eye movement, and appearance. HireVue removed facial analysis in January 2021 after external audits and ahead of the Illinois Artificial Intelligence Video Interview Act. Advice built on the old model (smile more, hold eye contact for the algorithm) is not what moves the score today. Camera setup still matters for the human reviewer who watches the shortlist, which is a different reason to get it right.

Accent and vocabulary size, by design. HireVue's published position is that its assessments are built by in-house industrial-organizational psychologists with adverse-impact testing, and that the models are trained to score competency evidence rather than dialect. Whether any given employer's configuration lives up to that is not something a candidate can verify, which is one reason Illinois and Maryland require disclosure and consent before an AI-evaluated video interview.

• Browser focus. Tab-switching during the recording is tracked. Reading a script off a second window is the classic way to trigger it.
• Scripted or AI-generated delivery. Answers that read as recited, or that cluster suspiciously close to other candidates' answers to the same question, get flagged for human review.
• Audio anomalies. Synthetic audio and heavy post-processing are detectable. Use the built-in recorder as intended.

The practical rule: prepare one story per competency, rehearse it until it is yours, then put the notes away before you press record.

The score is a sort order, not a verdict. Recruiters open the top of the ranked list first, watch the recordings, and decide who advances. That means two things for you: a strong transcript gets you watched, and the recording itself still has to hold up when a person plays it. Camera at eye level, light on your face, and answers that end on a result rather than trailing off all matter at that second stage.

The main HireVue guide covers the technical setup checklist and the 5-day plan; HireVue interview questions covers the three question formats with worked answers.