What are Asynchronous Video Interviews?
In August 2023 we recorded a set of asynchronous video interviews ourselves, on a platform called Interviewer.AI, using invented applications, to find out what the scoring actually rewards. One of them came back with 52 per cent for dress.
Here is what we found, what has changed since, and what Australian law will — and will not — require anyone to tell you.
What an asynchronous video interview is
An asynchronous video interview, or AVI, sometimes sold as an AI interview or a one-way interview, is a recording rather than a conversation. The employer sets the questions in advance. You record your answers alone, into a camera, usually against a timer of about a minute per question and often with a limited number of takes. Nobody is watching while you do it. The recording is then scored — sometimes by software, sometimes by a person reading a software-generated summary — and passed on, or not.
It sits between the application and the first real interview, and its purpose is to reduce a large pile of applicants to a small one without anyone’s diary being involved.
What happened when we recorded them
One recording, made deliberately casually — a ponytail, a gingham scarf, a collarless top — came back with 52 per cent for dress and 39 per cent for presentation.
We also scored badly on a category the platform called Sentiment, which was the correct result: the applications were make-believe and the enthusiasm was fake. Whatever else these systems measure imperfectly, one of them was measuring conviction, and it caught us.
The numbers are three years old. What they describe is not. As at August 2026, Interviewer.AI’s own published score breakdown still lists Professionalism among its five scoring areas, and still divides it into Dressing — “how formally you are dressed for the interview” — Grooming, “well-groomed and presentable” — and Body Language: “eye contact, posture, and use of hand gestures”. A second category, Sociability, includes “facial expressions and emotional openness”. The company describes 41 separate models measuring observable features including eye contact, emotional state and energy level.
We were marked on our scarf, and the marking scheme is still published.
The part that should give you pause
In January 2021, HireVue — the largest platform in this market, and the one Angelo Capuano reported in The Conversation in December 2023 was in use by more than a hundred employers including Telstra — announced it would stop using visual analysis in its pre-hire algorithms altogether. Its stated reason was not ethics but accuracy: appearance had far less correlation to job performance than language, and adding visual data to language data produced no extra predictive power. The company said it hoped the decision would become an industry standard.
It did not. Five and a half years later, a competitor in the same category still scores how formally you are dressed and how well groomed you look.
Both positions are published by the companies themselves. They cannot both be right about whether appearance predicts performance. And when an employer sends you a link, you are not told which of the two you are about to record into, what it weighs, or whether anyone has ever tested it against Australian candidates.
What we said in 2023, revisited
We speculated then that these systems are trained largely on US data, and that Australian reserve would score badly against them. We flagged at the time that we had no evidence. Some has since accumulated, and it points at something more specific — and more mechanical — than we guessed.
Interviewer.AI is a Singapore company. That tells you very little about what its models learned from, because almost nobody builds a speech recogniser from scratch. Vendors build on the same English-language stack everyone else uses, and that stack is measurably better at American English than at ours. Testing OpenAI’s Whisper across accent groups, Graham and Roll found Australian English produced significantly higher error rates than American English (JASA Express Letters, February 2024). Across six commercial recognisers — Amazon, Apple, Google, IBM, Microsoft and Kaldi — Tadimeti, Georgila and Traum measured non-American accents running roughly 16 to 49 per cent worse in relative terms than General American. That second study did not test Australian English; the Whisper study did, and found the same direction of effect.
This matters more than it looks. Three of the things Interviewer.AI scores — Articulation, Grammar and Sentiment — are language scores, and a language score is computed from a transcript. If the transcript is wrong more often for an Australian voice than an American one, every score built on top of it inherits the error. You are not being marked down for how you speak. You are being marked down for what the machine thought you said.
There is an irony in it worth sitting with. HireVue’s stated reason for switching the camera off was that language carries the predictive weight, not appearance. If that is right, and the language layer is the accent-sensitive one, then turning the camera off concentrates the problem rather than solving it.
No vendor in this market publishes its error rates broken down by accent. Nobody in Australia is required to ask for them.
Where the law actually stands
Three dates matter, and they do not point the same way.
December 2025. The Federal Government released the National AI Plan on 2 December 2025 and abandoned its earlier intention to legislate mandatory guardrails for high-risk AI. AI in hiring is now governed by the laws already on the books rather than by anything purpose-built.
Now. There is no general requirement for a private-sector employer to tell you that AI was used to assess you, or to explain how. Anti-discrimination law still applies — but as JobWatch put it in February 2026, an applicant has to prove discrimination occurred, and with an algorithm in the way, working out how and why a rejection happened becomes significantly harder. Applicants also sit outside most workplace protections, which are built for employees.
10 December 2026. From that date, Australian entities that use personal information in automated decision-making capable of affecting your rights or interests must say so in their privacy policy, including the kinds of information used and the kinds of decisions made. It does not give you a right to an explanation of your own result, and it does not stop the practice. What it does mean is that from December, an employer running an AI interview should be saying so in writing, somewhere you can read it.
Until then, the honest position is that you are being scored by a system nobody is obliged to describe to you.
How to record one well
Write the answers first. The format does not reward spontaneity. You will have roughly a minute per question and little or no opportunity to restart, so draft responses to the questions that always come — your interest in the role, what you would add to the organisation — and anchor each one in a specific example.
Then cut them. A minute is short. One example per answer, delivered steadily, beats four ideas at speed. The most common failure we saw was trying to fit an entire career into sixty seconds.
Rehearse out loud, not in your head. Reading silently tells you nothing about whether you can say it in the time.
Dress as you would for the interview this replaces. A plain collared shirt, a jacket, hair that stays put when you move your head. You should not need a stylist. Assume appearance is being scored — on at least one platform it demonstrably is, under the heading Professionalism, and you will not be told in advance which platform you are on.
Mean it. Sentiment was the one category that read us accurately, and it is not incidental: Positive Attitude is one of the five scoring areas, and sentiment appears again inside two of the others. Enthusiasm you do not feel is detectable. If you cannot summon any for the role, that is information worth acting on yourself.
Find out whose software it is. If the platform is named in the invitation, look it up. Several publish their scoring categories openly, and reading them takes five minutes. Knowing whether appearance is on the list changes how you prepare.
And keep the written application good
AVIs are becoming a routine screening step, not a replacement for everything else. Hybrid processes are the norm: the video gets you past the first cut, and the written material still does the work of making the case. Both have to be right.
If you want to practise the format before it counts, our executive interview coaching covers it.






















































