One-way video interviews: what they solved, and what they broke
One-way video interviews solved a real problem — recruiter hours consumed by first-round phone screens — by moving the cost onto candidates. That trade worked while candidates had no alternative. It works considerably less well now.
- Asynchronous video removed the scheduling bottleneck, which was a genuine and large win for volume hiring.
- It moved effort to the candidate: record, re-record, no interaction, no chance to ask anything.
- Video introduces assessment inputs — appearance, background, camera quality, accent — that have nothing to do with the job.
- Emotion and facial inference in hiring is prohibited for high-risk systems under the EU AI Act and is a discrimination exposure in Australia.
- The useful successor keeps the asynchrony and the structure, drops the video from the assessment, and gives the candidate an interaction rather than a recording.
The problem it genuinely solved
A phone screen costs a recruiter roughly thirty minutes including scheduling, no-shows and notes. At four hundred applicants that is two hundred hours, which nobody has. So the pipeline gets cut on CVs first — the least predictive evidence in the process — and the recruiter screens the survivors.
One-way video changed the arithmetic. Candidates record answers to fixed questions on their own time; the recruiter reviews at double speed, skipping ahead. Review time per candidate falls to a few minutes, which means far more candidates can be reviewed. That is a real improvement in coverage and it is why the format spread so fast.
What it cost
Completion
Asynchronous video has a well-known drop-off problem. The candidate is asked to perform to a camera, alone, with no interaction and often a countdown timer, before anyone from the company has spoken to them. Strong candidates with other options are precisely the ones most likely to decline — the format selects hardest against the people you most want.
Fairness
Video puts things in front of the assessor that the job does not require: the room behind the candidate, the quality of their camera and lighting, their internet connection, their appearance, their accent. Each of those correlates with socioeconomic background, national extraction or disability. Under Australian law those correlations are not an abstract concern — under s.360 of the Fair Work Act it is enough that a protected attribute was among the reasons.
Signal
A recorded monologue answering a fixed question is a weaker sample than a conversation. There is no probing, so a thin answer stays thin, and the assessor cannot distinguish a candidate who lacks the competency from one who misread the question. Fixed questions also mean the interview cannot go deeper where the evidence is interesting.
Comparing the formats honestly
| Phone screen | One-way video | Structured AI interview | |
|---|---|---|---|
| Scheduling cost | High — the binding constraint | None | None |
| Candidate effort | Low | High — recording, re-recording, no interaction | Moderate — a conversation, on their schedule |
| Probing on thin answers | Yes, if the recruiter thinks to | No | Yes, rule-governed and equal across candidates |
| Consistency across candidates | Low | High on questions, low on assessment | High on both |
| Assessment inputs | Voice and content | Content, appearance, setting, accent | Transcript content only, where the vendor is built that way |
| Candidate can ask questions | Yes | No | Depends on the platform |
| Evidence behind a score | Recruiter notes | Reviewer notes | Verbatim, timestamped citations |
What to use instead
The design goal is to keep what one-way video got right — no scheduling, every applicant assessed, consistent questions — and fix what it got wrong.
- 01Make it a conversation, not a recording. Adaptive follow-ups mean a thin answer gets probed rather than scored down, which is both fairer and better signal.
- 02Let the candidate choose the modality. Voice, video or text, in the language they are strongest in. Modality should not be an assessment input.
- 03Score transcript content only. No video frames, no audio features, no inferred demographics. This is an architectural choice, not a setting.
- 04Give the candidate something back. Personalised feedback for every applicant converts the format from an extraction into an exchange, which is most of what fixes completion.
- 05Require evidence for every rating, and permit abstention. A score without a citation is an opinion with a number attached.
If you are currently running one-way video and weighing a change, our comparison of AI interview platforms in Australia sets out how the main options differ on exactly these axes.