Recorded and machine-scored interviews

Three different things wearing one name

The one-way video interview. You are sent a link, questions appear on screen, you record answers with a time limit and often one retake. HireVue, Spark Hire, Willo and others. There is no interviewer.

The chat screen. A bot asks knockout and basic screening questions in a messaging interface before a human is involved.

The live AI interviewer. Newer, less common, a synthetic voice conducting a conversation. Treat it as a one-way interview that interrupts.

All three are used for the same reason: a posting with 1,400 applicants and one recruiter. Knowing that is useful, because it tells you the stage’s purpose is elimination, not selection.

What is actually scored, and what is no longer

Most systems transcribe your answers and score the text against a rubric of competencies: structure, relevance, whether you gave a specific example. Some also score pace and pauses.

Automated facial analysis was heavily marketed around 2018 to 2020 and has substantially retreated. HireVue publicly stopped using facial analysis in its assessments in 2021 following criticism and a complaint to the US Federal Trade Commission, and the research underpinning inferences from facial expression is contested. Some vendors still market variants of it. You cannot tell from the outside which you are facing.

The evidence base for the whole category is weaker than the marketing implies. Vendors publish validity studies; independent replication is thin. Treat confident claims in either direction with suspicion, including the claim that these systems are worthless: a structured, rubric-scored transcript is not obviously worse than a distracted human, and that is an uncomfortable thing to sit with.

The rules, which are real and moving

• Illinois requires notice, an explanation of how the AI works, consent, and deletion on request for AI-analysed video interviews.
• New York City requires bias audits of automated employment decision tools and notice to candidates.
• The EU AI Act classifies recruitment and selection systems as high risk, with obligations phasing in.
• Elsewhere, frequently nothing specific yet.

You are usually entitled to ask what is being assessed, whether a human reviews the result, and how long the recording is kept. Ask the recruiter by email. The answers are informative regardless of content.

How to do well in a recording

The reason people underperform is mechanical. Human conversation runs on feedback (a nod, a follow-up, a laugh) and with none of it, people speed up, wander and stop early. Compensate deliberately:

• Announce the structure. “Three things happened. First…” A rubric looking for structure now has it, and so does the human who skims the transcript.
• Ninety seconds per answer. Two minutes at most. Watch the timer once, then ignore it.
• Answer the question in the first sentence, then give the example. Do not build to a conclusion; there is nobody to hold.
• Look at the lens, not at yourself. Move your own preview window up near the camera if you can.
• Use the practice question properly. Almost every system offers one. Record it, watch it back, fix the audio and the framing before you start.
• Sound quality beats picture quality. Wired earphones with a microphone, a quiet room, and a hard surface behind you rather than a curtain. If the transcript is wrong, the score is wrong.
• Plain background, decent front light. A window in front of you, not behind.

Prepare four stories in advance (a conflict, a failure, a thing you shipped, a time you changed your mind) and expect to reuse them here.

Ask for what you need

If you have a speech difference, a disability, unreliable internet, or no quiet space, say so and ask for an alternative. In many jurisdictions employers have an obligation to consider reasonable adjustments, and a good number of companies will offer a live call instead without any obligation at all. The request is normal and is not held against you by anybody you would want to work for.

The limitation

These systems are documented to work less well for people with strong regional accents, atypical speech, poor bandwidth or noisy homes, which describes a substantial share of the people this course is written for. That is a genuine fairness problem and it is not yours to solve at 9pm the night before.

Two practical responses. Control what you can: audio, light, structure, timing. And weight the channel accordingly: a company whose entire first stage is an unexplained one-way recording has told you something about how it treats candidates, and that information is worth having early.