A practical guide to the AI interview: what it actually assesses, the three formats you'll meet, and how to prepare so an algorithm doesn't cost you a job you'd otherwise win.
You open a scheduling link expecting a call with a recruiter. Instead, you get a webcam prompt, a timer, and a synthetic voice asking you to describe a time you handled conflict at work. No human on the other end. Just a camera, a clock, and an algorithm scoring your answer.
This is the AI interview, and it's no longer a fringe experience. A Greenhouse survey reported by Fortune found that 63% of US job seekers have now been interviewed by AI at some point in their search. That's a majority of the market, not an edge case.
The same survey found real friction with the format: 38% of job seekers admitted to withdrawing from a hiring process specifically because it required an AI interview, and of those who stuck it out and completed one, 51% were later ghosted or are still waiting to hear back. Those numbers say something honest about where trust currently sits. But they don't mean you should refuse every AI interview that lands in your inbox. In a lot of cases it's simply the first gate a company uses before a human ever looks at you, and treating it as a throwaway step is how good candidates get filtered out before anyone real sees their work.
This guide is about what an AI interview actually assesses, the three formats you're most likely to run into, how to prepare for each one properly, and what you're reasonably entitled to ask for when a company puts an algorithm between you and the next stage.
What an AI Interview Actually Is
"AI interview" gets used loosely, so it's worth being precise. It generally means one of three things:
An asynchronous video interview where you record answers to preset questions on your own time, and either an algorithm or a human reviewer (or both) evaluates the recording later.
A chatbot or text-based screening where a conversational AI asks questions in a chat window and you type or speak your responses in real time.
A live voice interview where a synthetic voice agent conducts something closer to a real-time conversation, asking follow-up questions based on what you say.
What connects all three is that no human is present during the interaction itself, even though a human is very likely to review the output afterwards. That distinction matters more than most job seekers realise: the AI stage usually isn't the final decision. It's a filter, similar in spirit to an ATS scan of your resume, except this time it's evaluating how you speak rather than how you're formatted on paper. We've written before about how well employers can actually detect AI in your materials, and the same asymmetry shows up here: the company is comfortable using AI to evaluate you, but the interaction still feels, to the candidate, like being judged by a machine with no room for nuance.
That asymmetry is a large part of why trust in the format is shaky. It's also why this sits inside a bigger pattern we've covered in our piece on the AI hiring arms race: employers deploy AI to manage volume, candidates feel the process becoming more opaque and adversarial, and both sides escalate. The AI interview is one of the most visible flashpoints in that arms race, because unlike a resume scan, you can actually feel yourself being evaluated by it in real time.
The Numbers, in Context
It's worth sitting with what the Fortune/Greenhouse data actually says before deciding how to feel about AI interviews.
Read those three numbers together and a pattern emerges. AI interviews are now common enough that avoiding them entirely isn't a realistic strategy for most job seekers. But a meaningful share of candidates are opting out anyway, and even the ones who push through often don't get closure. That's not necessarily a reason to fear the format. It's a reason to treat it seriously and go in with a plan, because the companies using it well and the companies using it badly are both hiding behind the same interface.
What AI Interviews Are Actually Assessing
The uncomfortable truth is that most AI interview tools are not primarily judging your answer's content. They're judging a mix of signals, and the weighting varies wildly by vendor:
Keyword and concept coverage. Did you mention the specific skills, tools, or concepts the question was designed to surface? This is the closest thing to a resume keyword scan, applied to speech.
Structure and completeness. Does your answer have a clear beginning, middle, and end? Vague or rambling answers score worse even when the underlying experience is strong.
Pacing and fluency. Filler words, long pauses, and speaking rate are commonly scored, sometimes explicitly, sometimes as a byproduct of transcription confidence.
Sentiment and tone. Some tools attempt to detect enthusiasm, confidence, or defensiveness from your voice or word choice. This is the most contested part of the technology, and also the part most likely to misfire on non-native speakers or anyone who doesn't perform enthusiasm the way the model expects.
Relevance to the role. Whether your example maps clearly onto the competency the question was written to test (leadership, conflict resolution, ownership, and so on).
None of these are mysterious once you know they're being measured. They're also not that different from what a well-trained human interviewer is listening for, they're just being measured more literally and less forgivingly. A human interviewer might overlook a rambling answer if your body language and follow-up questions show genuine competence. An algorithm scoring a transcript has no such context, which is exactly why structure and directness matter more here than they do in a live conversation with a person.
The Three Formats, and How Each One Actually Plays Out
Format 1: Asynchronous Video Interview
You get a link, a list of questions (usually 4 to 6), and a limited time to answer each one, often with one or two attempts and a countdown timer visible on screen. There's no interviewer to read your face or nod along, so the entire evaluation rests on the recording.
If your invite is from HireVue, our HireVue interview guide covers what its AI scores, according to HireVue itself, and 25 common questions.
What trips people up: the countdown timer causes people to rush, and the lack of any human feedback (no nodding, no "mm-hmm") makes the silence feel much longer than it is. Candidates either speed through answers and skip detail, or freeze and burn their limited retakes.
How to prepare: Record yourself answering common behavioural questions on your phone before the real thing. Watch it back once, not to be your own critic, but purely to check pacing. Most people are shocked at how much faster they actually talk than it feels. Practising with a real camera, even a phone propped against a mug, removes 80% of the awkwardness on the day. Our AI mock interview includes a timed AI first round, with questions taken from the job ad.
Format 2: Chatbot or Text-Based Screening
A conversational interface asks a series of questions, sometimes branching based on your previous answer, and you respond by typing or occasionally speaking into a text field. These are frequently used as a first-pass screen before any human or even any video round, often replacing what used to be a 15-minute phone screen with a recruiter.
What trips people up: typing under a soft time pressure encourages short, generic answers, and the branching logic means an early vague answer can lead you down a shallower line of questioning that never lets you show your best material.
How to prepare: Treat every chatbot answer like a resume bullet point rather than a chat message. Lead with the outcome, then the context, briefly. If the tool gives you the option to elaborate or add a follow-up, take it. Specificity is what separates a real answer from filler, in exactly the same way it separates a strong resume bullet from a vague one.
Format 3: Live Voice AI Interview
This is the newest and most conversational format: a synthetic voice conducts something close to a real dialogue, asking clarifying or follow-up questions in response to what you say, sometimes adaptively probing deeper on a topic you mentioned. It's designed to feel like talking to a person, and for a lot of candidates that's precisely what makes it unsettling.
What trips people up: people either overcorrect by talking to it like a search engine (short, keyword-heavy fragments) or undercorrect by rambling the way they might in a genuinely relaxed human conversation. Both underperform against a clear, complete answer delivered at a normal conversational pace.
How to prepare: Speak to it exactly as you would to a competent but literal-minded interviewer: complete sentences, a clear structure, and enough context that the answer would make sense to someone hearing it cold. Don't try to guess what "makes AI happy." Just answer the actual question well.
The STAR Drill That Works for All Three
Every AI interview format, regardless of vendor, tends to reward the same underlying structure: a clear situation, a specific action you took, and a measurable result. This is the STAR method, and it happens to map almost perfectly onto what these tools are built to detect, because structured answers are easier for a model to parse and score than a stream of consciousness.
Before any AI interview, prepare 5 to 6 stories from your actual experience, each mapped to a common competency: a time you solved a problem under pressure, a time you disagreed with someone and how you handled it, a time you took ownership of something that went wrong, a time you influenced an outcome without formal authority. Write each one down in three or four sentences, practice saying it out loud twice, and you'll have material ready for almost any question an AI interview throws at you, without sounding rehearsed.
This is also good practice for classic human interviews, not just AI ones. If you've ever frozen on "walk me through your resume," the underlying fix is the same: know your own story well enough that you can tell it in whatever shape the question demands. Our guide on answering "walk me through your resume" is worth reading alongside this one, since a lot of AI interview opening questions are just a variation on that same prompt.
Practical Prep Checklist
A short list you can actually run through the night before:
Test your setup first. Check your camera angle, lighting, and microphone before the real session starts. Most tools give you a practice run. Use it, even if it feels unnecessary.
Sit somewhere quiet with a plain background. Not because the AI is scoring your decor, but because background noise degrades transcription quality, and a poor transcript is a worse foundation for scoring than a strong one.
Have your STAR stories ready, not scripted. Memorising a script word for word makes you sound stilted and increases the odds you freeze if the AI's follow-up doesn't match what you rehearsed. Know the shape of the story, not the exact sentences.
Speak at a normal, complete pace. Don't rush to beat a timer, and don't pad with filler to fill time. Aim to answer the actual question, then stop.
Treat every question as one competency, not a trick. AI interview questions are almost always mapped to a specific skill the employer wants evidence of. Answer the competency directly rather than trying to be clever or unconventional.
Reread the job description right before you start. Many AI interview tools weight relevance to the role, so having the language from the posting fresh in your mind helps you naturally use terms that map to what's being scored.
Your Rights, and What You Can Reasonably Ask For
An AI interview is still part of a hiring process a company is running, and you're allowed to ask sensible questions about it, the same way you'd ask about salary range or team structure. None of the following are confrontational, they're basic due diligence:
Ask whether a human reviews the results. A simple, polite question to a recruiter ("does someone on your team review these, or is the scoring fully automated?") tells you a lot about how seriously the company treats the tool.
Ask what happens if there's a technical issue. Frozen video, a dropped connection, or a misfire on a retake attempt shouldn't cost you the role. Most reputable companies have a fallback, but it's worth confirming before you start rather than discovering there isn't one.
Request an alternative format if you have a disability or access need. Voice-based and video-based assessments can disadvantage candidates with speech differences, hearing impairments, or certain neurodivergent conditions. A reasonable employer should be able to offer an alternative, and asking is a legitimate accommodation request, not a special favour.
Ask for feedback if you're rejected after completing one. You may not get it, but asking costs nothing, and given how many candidates report being ghosted after an AI round, a direct request for feedback occasionally gets a genuine answer where silence otherwise would have followed.
Expect a timeline, and follow up once if you don't get one. Given how common the ghosting pattern is after these interviews, it's reasonable to ask upfront when you should expect to hear back, and to send one polite follow-up if that date passes.
None of this guarantees a better outcome. But it shifts the interaction from "algorithm judges you in silence" to "you're a participant in a process you understand," which is a meaningfully different experience even when the result is the same.
A well-prepared answer only lands if the story behind it is actually strong on paper too. JobSprout's AI assistant can help sharpen the same achievement stories you'll use in an AI interview into concrete CV bullet points, so the language you practice saying out loud matches the language a recruiter already read about you.
Frequently Asked Questions
Is it fair for a company to make me do an AI interview?
Fairness depends heavily on execution. An AI interview used as an early, low-stakes screen with human review afterwards is a reasonably defensible way to manage volume. An AI interview used as the sole basis for a rejection, with no human oversight and no path to feedback, is far harder to defend. The Fortune/Greenhouse data showing 38% of candidates withdrew suggests a large chunk of the job-seeking public has already made up its mind that the current implementation often isn't fair. You're entitled to that opinion too, and to act on it by asking questions or declining, but for most people it's still worth doing well rather than opting out of a majority of the market.
Can I tell if the interview is actually being scored by AI versus just recorded for a human to watch later?
Not always from the interface alone. Some tools clearly disclose that scoring is automated; others are vaguer. If it's not obvious, ask the recruiter directly. It's a completely normal question, and a company that can't answer it clearly is telling you something about how much oversight exists internally.
Should I use AI to help me prepare answers, or does that feel hypocritical given the other side is using AI to judge me?
Using AI to prepare is not the same as using AI to fabricate. Practising your delivery, refining how you phrase a real achievement, or getting feedback on whether a story is clear are all legitimate uses of AI as a coaching tool. The distinction that matters, in interviews and on your CV alike, is whether the substance is genuinely yours. An AI interviewer asking about a project you actually led is a very different situation from AI generating a project you never worked on.
What if the AI interview asks something I've never practiced?
Fall back on the STAR structure rather than the specific story you rehearsed. Pick the closest real example from your experience, even if it's not a perfect match to the question, and walk through situation, action, result in that order. A slightly imperfect but well-structured real example consistently scores and reads better than a vague, meandering answer searching for the perfect anecdote.
Does a bad AI interview score mean I'm out of the running?
Not necessarily, and this is genuinely important context. Many companies use the AI stage as one signal among several, not a hard cutoff, and plenty of hiring managers manually review borderline scores before rejecting anyone. That said, if you consistently find AI interviews going badly across multiple applications, it's worth deliberately practising the format itself (timing, pacing, structure) rather than assuming the content of your experience is the problem.
Are AI interviews going to replace human interviews entirely?
Unlikely, at least for the foreseeable future, and the ghosting pattern in the Fortune/Greenhouse data is itself a warning sign for companies relying on AI interviews too heavily: candidates who feel unseen by a process disengage from it, including strong candidates a company would want to keep. The more durable pattern is AI handling early-stage screening at scale while final decisions stay with humans who actually meet you, which mirrors the broader shift we cover in our piece on the AI hiring arms race.
The Bottom Line
An AI interview is a structured filter, not a trick question dressed up as technology. It rewards the same things a good human interviewer rewards: clear, specific answers about real experience. The format changes how that gets delivered, not what gets evaluated underneath it.
Prepare your stories, test your setup, ask reasonable questions about the process, and treat the algorithm the way you'd treat any interviewer who's listening carefully but has no patience for vagueness. That's a skill worth having regardless of who, or what, is on the other end of the call.
If you want your CV to hold up the same story you tell in an AI interview, JobSprout can help you turn your real achievements into sharp, specific bullet points that a recruiter, and an algorithm, can both follow.
