Skip to main content
Back to Blogs
Interview Prep7 min read

Recorded Video Interviews: What Is Actually Scored, and What Is Not

HireVue and Willo record what you say, not your face. What reviewers mark, how think time and retake rules differ, and how to structure an answer alone.

March 2020
when HireVue switched off facial analysis
Under 0.25%
what it said visual data contributed to its models
10 days
notice New York City requires before an automated tool scores you
30 days
to delete your recording on request under Illinois law

Nobody is on the other end while you record. What the employer receives is a file, usually with an automatic transcript attached, opened days later in a batch of forty. Most of what goes wrong in a one-way interview comes from preparing for a conversation and then delivering into something much closer to a submitted document.

What is actually at the other end

Recorded interviews reach a reviewer in one of three ways, and the invitation rarely tells you which. A recruiter may watch every answer against a written scoring guide. A scoring model may run over the transcript first and rank the batch, with humans watching only the top of the pile. Or the two run together, with the model sorting and a hiring manager confirming. In every version, the thing being judged is a recording made without feedback, by someone who cannot ask you a follow-up.

HireVue, Willo, Spark Hire and Talview are delivery platforms rather than test publishers. The employer writes the questions, sets the clocks, and decides what happens to the recording. That matters for preparation: there is no fixed question bank to learn, and the rules you are sitting under were chosen by the company, not the software.

Your face is not the thing being read
HireVue removed facial analysis from its video interview assessments, announcing it in January 2021 and saying the feature had been switched off in March 2020. Its stated reason, reported alongside an audit of its algorithms, was that visual data added under a quarter of one percent to what its models predicted once language analysis improved. Some vendors do still measure speech features such as pace and pausing, and none publish the exact list. The practical conclusion holds either way: expression management is not a lever you can pull, and candidates who try to pull it look glassy for the full two minutes.
Where the prep hour usually goes
Mostly invisible in the file
Smiling on cue and worrying about eye movement
Rearranging the shelf behind the chair
Memorising a first sentence word for word
Deciding whether a jacket is required
Reading forum threads about what the AI detects
What survives into a transcript
What a reviewer can mark
Whether you answered the question that was asked
Verbs attached to you rather than to the team
One number, date or quantity per answer
Whether the result arrived before the cut-off
Whether five answers describe five different situations

Think time and retakes are settings, not facts about the platform

Candidates swap advice as though every HireVue invitation behaves the same way. It does not. Think time, the recording cap and the retake allowance are configured per job, so the person who told you they had three attempts may have been sitting a different configuration of the same product.

Willo is a useful example of how much the mechanics can differ. Its timer is a guard rail rather than a gate: the clock counts up rather than down, it keeps running if you close the browser, and the employer sees afterwards whether you came in under or over the time they allocated. That clock also includes any retakes you use. An employer can set retakes to zero, in which case your first take uploads automatically, and the screen shows you how many you have left. So on a platform where the timer counts up, a third attempt is not free. It is visible.

Three things to read on the instruction screen
Before question one, every platform states the rules for that job. Note whether the timer counts down to a hard stop or counts up as a record of how long you took, how many retakes you have per question, and the maximum length of a single answer. Those three facts decide your entire risk appetite, and they are on a screen most people click through in four seconds.

Take the unscored practice question if one is offered. It is there to check your levels, and it burns off the first take, which is almost always the worst one. Where retakes exist, keep them for a genuine failure, a lost thread or a truck outside, rather than for polish. Second and third attempts usually get flatter, not better, because you start reciting the version you just said.

Structuring an answer when nobody is nodding

In a live interview an interviewer rescues you. They nod, they interrupt with the question you should have been answering, they let you restart. On camera, an answer that wanders for two minutes is submitted exactly as it wandered. The fix is to build the shape into the first ten seconds instead of hoping to find it as you go.

Headline: the outcome and the competency in one sentence0:00 to 0:10
Situation: two sentences, no more0:10 to 0:25
Action: your decisions, in order, in the first person0:25 to 1:10
Result: the number, then what changed afterwards1:10 to 1:25
Landing sentence, rehearsed1:25 to 1:30

Three devices do most of the work in that shape. Lead with the outcome, because a reviewer skimming at speed decides in the first fifteen seconds whether to pay attention. Cap the background at two sentences, because setting the scene is what eats the clock. And signpost out loud: saying "I did three things" and then counting them off makes the structure audible to someone who is listening rather than reading.

Write out exactly two sentences in advance: the one you open with and the one you stop on. Scripting the middle makes you sound read. Scripting the ending is what stops you trailing off into silence and then apologising into the lens.

The four ways candidates lose this stage

Failure mode
The clock ends before the result does
The single most common loss. Use a halfway rule: when half the time has gone, if you have not yet said what you decided, jump straight to it. A thin situation with a clear outcome scores; a rich situation cut off mid-sentence does not.
Failure mode
Reading a script off a second screen
The eyeline drifts, the pace flattens, and written syntax shows up in the transcript as well as on the video. Keep three anchors on a card taped beside the lens: the situation in two words, the decision, the number.
Failure mode
Dead-eyed delivery at a lens
Your eyes follow your own face on screen, which reads as looking away for the whole answer. Move the self-view window directly under the camera or cover it, pick one real person to address, and speak with slightly more energy than feels natural. Recordings flatten energy.
Failure mode
Five answers, one story
Reviewers see the whole set together. Five variations on the same group project reads as one experience rather than five. Map your examples to distinct situations before you start and do not reuse one, even when it fits better.

Setup that measurably changes the file

Most setup advice is cosmetic. A few things genuinely alter what is recorded. Put the light in front of you, not behind: a window or a lamp facing you beats an expensive camera lit from the back. Get the lens to eye level, which for a laptop means a stack of books, because a camera looking up at you is the difference between a professional recording and a video call from a sofa. Sit close enough that your head and shoulders fill the frame, roughly an arm span from the screen.

Audio deserves more attention than it gets. Where the employer scores from an automatic transcript, anything that degrades the transcript degrades the input to the score. A hard room with bare walls, a fan, or a laptop microphone a metre away all produce transcription errors on exactly the specialist words you were hoping to be credited for. Wired earbuds with an inline microphone, a closed door and a room with soft furnishings will do more for you than a new webcam.

Do the whole thing on the exact device and browser you will use, close everything else so the upload has bandwidth, and record with days in hand. Invitations expire quietly, and every recoverable problem stops being recoverable at 11pm on the deadline.

What you are entitled to ask, depending on where you are

Rules on automated assessment sit with the employer rather than the platform, so the recruiter is the person to ask. What you can ask for varies by jurisdiction.

  • Illinois. The Artificial Intelligence Video Interview Act requires an employer to tell you before the interview that AI may analyse it, explain how it works and what characteristics it evaluates, and get your consent. On request, they must delete the recording and instruct anyone they shared it with to delete their copies, within 30 days.
  • New York City. Local Law 144 requires an annual independent bias audit of automated employment decision tools, a public summary of the results, and at least 10 business days of notice to candidates before such a tool is used on them.
  • UK, Netherlands and the rest of the EU. Under the GDPR, decisions based solely on automated processing that significantly affect you carry extra protections, including the right to ask for human involvement and to contest the outcome. In practice the useful question is simply whether a human reviewed your recording.
  • Australia.There is no dedicated statute for AI video interviews, so the handling of your recording falls under general privacy law and the employer's own policy. Asking what happens to the file, and for how long it is kept, is still reasonable.

Practise the format, not the answers

You cannot know the questions, because the employer writes them. What you can rehearse is the part candidates report as hardest: speaking fluently to a lens that does not respond, landing inside a clock, and stopping on purpose. That is a performance skill, and it improves quickly with honest review. A recorded mock interview with a prep window and a hard cap gets you most of the way there, and our guide to passing a video interview covers the question types you are most likely to meet. If your invitation came from HireVue, check which HireVue product you were sent before you spend the week preparing for the wrong one.

Ready to Start Practicing?

Apply these strategies with our comprehensive practice platform

Start Practising Free