Nine years of screening resumes taught me that most candidates are still preparing for the wrong interview. They spend the week before a screen practicing eye contact and posture for a human being who, increasingly, isn't there yet. By the time a live person actually looks at your file, an algorithm may have already ranked you.
That's not a scare tactic. It's just where hiring has moved. Roughly two-thirds of job seekers have now been through at least one AI-run interview, up 13 percentage points in just six months according to a [2026 Greenhouse survey](https://www.greenhouse.com/newsroom/63-of-job-seekers-have-faced-an-ai-interview-most-havent-had-a-good-one-yet) of nearly 1,200 candidates. If you're actively applying anywhere with a high-volume pipeline (retail, banking, customer support, entry-level corporate roles) you will run into one of these before your job search is over.
What an AI video interview actually is
Two very different things get lumped under "AI interview," and mixing them up is where most preparation advice goes wrong.
- 1Asynchronous, one-way video screening: you get a link, answer a fixed list of questions on camera within a time limit (often 30 to 90 seconds per answer, one or two takes), and submit it for later review. This is what HireVue, Spark Hire, and Modern Hire are best known for.
- 2Live AI-assisted interviews: a chatbot or voice agent asks questions in real time and adapts follow-ups based on your answers, closer to a live phone screen but with software on the other end instead of a person.
Both funnel into the same outcome: your responses get converted to text and scored against a model of what a strong answer looks like for that role.
How the scoring actually works
Here's the mechanic that surprises most candidates: modern AI interview tools mostly aren't looking at your face anymore. HireVue built its early reputation on exactly that: scanning facial expressions, tone, and word choice against 58 behavioral markers to predict job performance. Then an independent audit by O'Neil Risk Consulting found that nonverbal signals like expression and gaze contributed roughly [0.25% to the predictive score](https://fortune.com/2021/01/19/hirevue-drops-facial-monitoring-amid-a-i-algorithm-audit/), basically noise, while creating real bias risk. HireVue dropped facial analysis entirely in January 2021, and most serious competitors followed the same path under regulatory pressure.
What's left, and what actually drives your score, is natural language processing (NLP) applied to what you say. The system transcribes your answer, then compares it against a competency rubric built from patterns in how the employer's existing high performers answered similar questions in training data: specific examples, relevant keywords and skills, structured reasoning, and quantified outcomes. Mechanically, it isn't that different from how a resume ATS scores keyword overlap against a job description, just applied to speech instead of text.
What this means for you
You're not performing for a camera. You're generating a transcript that needs to contain the right substance, said clearly enough that a speech-to-text engine can capture it correctly.
The legal backdrop you're entitled to know about
If you're applying in New York City, there's a rule working in your favor: Local Law 144 requires any employer using an "automated employment decision tool," which explicitly includes AI video interview analysis, to commission an independent annual bias audit, publish a summary of the results, and give you at least 10 business days' notice before the tool is used to evaluate you, with the right to request an alternative process. Violations carry penalties starting at $500 per instance and climbing to $1,500 per day for continued noncompliance.
Even outside NYC, this matters because it tells you what a well-run hiring process looks like. About 70% of candidates in the Greenhouse survey said they weren't told an AI would be assessing them, and a fifth only found out once the interview had already started. If a recruiter can't tell you clearly what the tool measures, it's a fair question to ask before you invest 30 minutes recording answers.
How to actually prepare
- 1Treat the questions like a real interview, because they are one. Prepare with the STAR format (Situation, Task, Action, Result) exactly as you would for a human.
- 2Say the keywords out loud, not just on your resume. If the job description emphasizes specific skills or phrases, work that language into your spoken answer where it's genuinely true.
- 3Nail the take, because you usually only get one or two. Practice out loud beforehand with a timer so you're not structuring your answer live on camera.
- 4Fix the boring stuff first: quiet room, light source in front of you, camera at eye level, stable internet. A garbled answer transcribes badly regardless of what you actually said.
- 5Speak in complete sentences, at a normal pace. Rushed or mumbled speech mis-transcribes more often than clear, measured delivery.
- 6Don't perform for the camera. With facial and tonal analysis mostly gone from serious platforms, that energy is better spent structuring a stronger answer.
Before → After
Before: "I'm a really good communicator and I work well with others, especially under pressure." After: "In my last role, a vendor delay put a product launch two weeks behind schedule. I set up a daily 15-minute sync between engineering and marketing so both teams could adjust in real time, and we still shipped on the revised date with no scope cut." The second answer contains the concrete nouns, actions, and outcome an NLP rubric is built to reward.
If something feels off, ask
You're allowed to ask a recruiter what the tool measures, whether an alternative assessment is available, and how the results factor into the decision. A reasonable employer will have a straightforward answer. If they don't, that tells you something about the process you're walking into.
Key takeaway
An AI video interview is scored almost entirely on what you say and how clearly you say it, not on your face or your energy. Prepare STAR-format answers with real numbers, say the job description's keywords out loud where they're true, fix your lighting and audio, and treat your one recorded take like it's the only interview that matters, because for the purposes of getting to the next round, it is.