Recorded video interviews and AI scoring are the riskiest things in the whole toolkit. They are also genuinely useful in a few situations. The difference is between recording an interview and letting software judge one.
What "async" means
Asynchronous, usually shortened to async, just means not live. The candidate records answers to set questions on their own, and someone watches later.
Where it genuinely helps
- Time zones. When there are no working hours you share.
- Large numbers early on, where the alternative is a CV screen rather than a conversation. Then async is better than nothing.
- Flexibility for the candidate. People in work, with caring responsibilities, or who cannot take calls during the day often prefer recording at 9pm to asking for an afternoon off.
- Consistency. Everyone gets identical questions with no interviewer variation.
Where it hurts
- People drop out. Async stages lose candidates, and they lose the ones with other options first. Talking to a camera with nobody responding is unpleasant and easy to abandon.
- It excludes people. Speech differences, anxiety, being neurodivergent, poor internet, no private space to record. None of those relate to doing the job, and all of them make a recorded video go worse.
- It is one-way. The candidate gives you their time and learns nothing about you. Used early in a competitive market, it says your time matters and theirs does not.
- No follow-up questions. The probes from the last lesson — the part that tells doing apart from describing — are impossible.
AI scoring of video
This is where the warning gets specific. Some systems score recorded interviews by analysing facial expression, tone of voice, speaking speed and word choice, and inferring personality or ability from them.
Two things are worth knowing.
First, the link between how someone's face moves and how well they do a job is weakly supported. And scoring it disadvantages anyone whose expressions, accent, speech rhythm or fluency differ from what the system was trained on — which maps directly onto disability, being neurodivergent, and not speaking English as a first language.
Second, the industry itself has partly backed away. HireVue removed facial analysis from its assessments in 2021 after an external review and sustained criticism.
Analysing what someone said is a different matter. Scoring the content of an answer against your criteria is defensible. Scoring how their face moved is not.
The rules that apply
| Rule | Why it matters here |
|---|---|
| Illinois AI Video Interview Act | Requires notice, an explanation of what the AI judges, consent, and deletion on request |
| New York City Local Law 144 | Automated hiring tools need a yearly independent bias audit and notice to candidates |
| EU AI Act | Recruitment counts as high risk, with duties attached |
| Data protection rules | Rights around automated decisions; video is sensitive |
Module 11 covers these properly. The point here is that this is the single feature most likely to be specifically regulated where you operate.
A position you can defend
- Use async for practical reasons, not for judging — or where the alternative is no conversation at all.
- Score what was said against your criteria. Never score appearance, tone or manner.
- Always offer a live alternative, shown just as prominently, with no disadvantage.
- Tell candidates what is automated and what a person reviews.
- Have a person review every rejection that came out of an async stage.
One question for any vendor
"What parts of the recording does the score depend on?" If the answer includes anything about face, tone or speech patterns rather than the content of the answer, that is your answer — and it is the one you will have to defend.