AI video interview software: how it scores candidates and what to ask vendors

Three formats: recorded video, AI conversations and interview notes

In a one-way interview, the candidate sees a question on screen and records an answer on camera. HireVue, which calls this an on-demand interview, describes a 3-stage process in its explainability statement: speech is transcribed to text, the text is interpreted, and each answer is scored against competencies that the employer chooses with HireVue's industrial and organizational psychologists.

Conversational interviews replace the fixed script with a dialogue. HireVue now also sells an AI Interviewer that asks probing follow-up questions in a two-way conversation. Greenhouse's Voice AI runs a structured spoken interview that candidates complete on their own schedule, and Talview's Ivy calls candidates on an ordinary phone line and asks follow-ups based on their answers.

A third group records interviews that people run. These interview intelligence tools transcribe the conversation and draft notes: BrightHire captures AI-generated notes and has added an automated first-round interview, and Metaview writes notes and scorecards automatically. Zoom announced its acquisition of BrightHire in November 2025. Our directory lists 14 video and AI interview tools and 6 interview intelligence tools, with vendors based in the United States, South Korea, Japan, Australia, Spain and other countries.

Sources: [1][2][3][4][5][6][7][8]

How AI video interview software scores answers

The vendors we checked score content first. HireVue's explainability statement says its AI relies only on what the candidate says and does not assess facial expressions, body language, background or tone of voice. Greenhouse lists what its Voice AI does not score: accent, tone, fluency, pace, stutters, pauses, personality and confidence.

Scores are tied to rubrics that the employer controls. HeyMilo gives a score per question against recruiter-defined rubrics, with a written rationale based on what the candidate said. Sapia.ai runs its interview as a text chat of 5 questions and argues that written answers are a fair way to judge soft skills, behavioral traits and cognitive ability.

Not every market has moved the same way. ViewinterHR, from the South Korean company Genesis Lab, says it evaluates voice, answers, facial expressions and behavior together, and it holds a reliability certification from Korea's TTA. inAIR, another Korean tool, analyzed voice pitch, facial expressions and eye movements, according to a 2021 Korea Herald report that counted about 450 companies using it the year before.

How the score is used matters as much as how it is made. HireVue's statement shows candidates sorted into score tiers and leaves the decision on each tier to the employer, and Greenhouse says Voice AI never rejects or advances a candidate. harutaka, a Japanese interview platform, says in its AI policy that it does not build features that let AI decide pass or fail.

Sources: [1][3][9][10][11][12][13][14][15]

What vendors stopped scoring: faces first

Facial analysis was the first signal to go. HireVue says it removed the visual analysis component from all new assessments early in 2020, because advances in language processing meant visual analysis no longer added significant value. It announced the change on January 12, 2021, together with an audit by O'Neil Risk Consulting and Algorithmic Auditing, and SHRM reported that the feature had been discontinued in March 2020.

Independent tests showed why the content of an answer matters. In July 2021, MIT Technology Review reporters answered the questions of 2 English-language interview tools in German. Curious Thing still rated the candidate 6 out of 9 for English competency, and MyInterview, which analyzed intonation to infer personality, reported a 73 percent match for the job.

Some employers stepped back as well. Korea Airports Corporation stopped using AI interviews after 1 year, with officials telling the Korea Herald they were not sure exactly how and what the AI was evaluating.

In the EU, the law now draws the line. Article 5(1)(f) of the AI Act has prohibited AI that infers emotions in the workplace since February 2, 2025, except for medical or safety reasons, and the European Commission's guidelines on prohibited practices, C(2025) 5052, say the ban covers candidates during selection and hiring. In the United States, Maryland has required an applicant's signed consent since October 1, 2020, before an employer uses facial recognition during a job interview (Labor and Employment Article, section 3-717).

Sources: [16][17][18][12][14][15][19][20][21][22]

Consent and notice: what Illinois, New York City, the EU and the UK require

Illinois was first. Under the Artificial Intelligence Video Interview Act (820 ILCS 42/5), in force since January 1, 2020, an employer that uses AI to analyze recorded video interviews for positions based in Illinois must tell each applicant beforehand that AI may be used, explain how it works and what general types of characteristics it evaluates, and obtain consent. Section 10 limits who may receive the videos, and section 15 requires deletion within 30 days of an applicant's request, including copies passed to others.

Since January 1, 2022, section 20 has required employers that rely solely on AI analysis to decide who gets an in-person interview to report applicants' race and ethnicity to the state every year. The Department of Commerce and Economic Opportunity received no such data for the 12 months ending November 30, 2023, or for the 12 months ending November 30, 2024. Its reports do not say whether no employer lets AI decide alone or whether some do and do not report.

New York City's Local Law 144 of 2021 (Administrative Code sections 20-870 to 20-874), enforced since July 5, 2023, applies to tools that substantially assist or replace discretionary hiring decisions, which can include AI interview scoring. Employers need an independent bias audit from the past year, a published summary, and a notice to candidates who live in the city at least 10 business days before use; our guide to hiring bias audits covers the details. Illinois' Human Rights Act (775 ILCS 5/2-102) has also required notice when AI is used in hiring decisions since January 1, 2026, and Colorado's Senate Bill 26-189 will require notice at the point of interaction and an explanation of adverse outcomes within 30 days from January 1, 2027.

In the EU, AI that evaluates candidates is high-risk under Annex III, point 4(a) of the AI Act, and after the Digital Omnibus, Regulation (EU) 2026/1744, those duties apply from December 2, 2027. One rule already applies: since August 2, 2026, Article 50(1) has required providers to make sure people know they are interacting with an AI system unless that is obvious, which reaches AI interviewers. In the UK, section 80 of the Data (Use and Access) Act 2025, fully in force since February 5, 2026, allows solely automated significant decisions only with safeguards such as information, human intervention and a way to contest the decision.

Sources: [20][23][24][25][21][26][27][28][29][30][31][22][32]

Candidates are wary, while vendor surveys look positive

Public opinion in the United States is cautious. In a Pew Research Center survey of 11,004 US adults in December 2022, 71 percent opposed AI making final hiring decisions, 66 percent said they would not want to apply for a job where AI helps make hiring decisions, and 41 percent opposed AI reviewing job applications. In the same survey, 70 percent opposed employers using face recognition to analyze employees' facial expressions.

Vendors report more positive numbers from people who completed their interviews. HireVue cites feedback from 1.5 million candidates, 80 percent of whom said they enjoyed the experience and 89 percent of whom said it respected their time. Greenhouse reports an average rating of 4.5 out of 5 for Voice AI and an opt-out rate of 10 percent. Both sets of figures come from the vendors and describe people who took the interview.

Speech recognition does not work equally well for everyone. HireVue's test of its transcription provider found word error rates of 12 percent for speakers with a Canadian accent and 22 percent for participants from China, lower than the 17 to 45 percent it measured for alternative services. When scoring starts from a transcript, accent can change the input, so ask vendors for error rates by language and accent.

Sources: [33][1][3]

Questions to ask AI video interview vendors

Most tools in this category do not publish prices, so ask these questions before the first quote. In our directory, 11 of the 14 video and AI interview tools price on request, and 5 say they can host data in the EU.

Ask for documents, not demos. A vendor that scores candidates should be able to share its latest bias audit, a written description of the inputs its model uses, like HireVue's explainability statement, and accuracy figures for the languages it supports. Compare those answers across the AI interview tools in our directory before you shortlist.

Plan for records early. Illinois gives applicants the right to have their videos deleted within 30 days of a request, while Article 26(6) of the EU AI Act will require employers to keep a high-risk system's logs for at least 6 months once the duties apply on December 2, 2027. Your contract should cover both.

Sources: [8][1][20][34][26]