The AI interview: what the employer’s AI interviewer asks, how it scores you, and how to prepare

An AI interview is a first-round interview an employer runs with software instead of a recruiter. Three different formats go by the name, and they need different preparation, so check which one your invitation describes.

The conversational AI interviewer. A two-way interview by voice, sometimes with a video avatar. It asks questions built from the posting and your resume, follows up on your answers, times them, moves on at the limit and scores the transcript. HireVue’s AI Interviewer, Alex, Ribbon, micro1’s Zara and Mercor all run this format. Most of this guide is about it.

The recorded one-way video. A question appears on screen, a countdown runs, and you record your answer to nobody. HireVue’s candidate FAQ says its recorded assessments have between five and eight questions. Candidate guides report about 30 seconds to prepare and up to three minutes to answer, but the employer sets the clock, so read your own screen. There are no follow-ups.

The text chat screen. A chat by text message or on the careers site. Paradox’s Olivia asks a handful of yes-or-no questions on authorization, shifts, transport and start date, then offers an interview slot. Sapia asks a few written questions with no timer. These need honest answers more than rehearsal.

The vendors with a conversational interviewer include HireVue (whose AI Interviewer launched in June 2026), Alex, Ribbon, micro1, Mercor, Hubert and HackerRank. Employers rarely announce which vendor they use, so most of what is public comes from vendor case studies, press and candidate reports. Thita’s list of companies using AI interviewers, dated October 2026, is the most careful compilation we found. From it:

• Zapier piloted the Ezra interviewer: about 250 interviews of 15 to 20 minutes, opt-in, with a recruiter reviewing every result.
• ManpowerGroup and Experis announced a partnership with Hubert in March 2026.
• Snowflake, Snorkel and Capgemini tested HackerRank’s AI interviewer before it became generally available on October 5, 2026.
• Mercor and micro1 use their own AI interviewers as the screen for their own listings.

LinkedIn began testing AI screening interviews in its small-business hiring product on March 12, 2026. The hirer must review the AI’s output before deciding. Candidate reports add more names, such as a roughly 15-minute AI round at Tesla in September 2026, where the recruiter offered a traditional screen instead. Treat a single report as one person’s experience, not the employer’s policy.

The vendors publish less about their questions than about their scoring, but their pages and candidate accounts agree on the shape. Your interview may skip a step or change the order.

1. 1The introduction. It says it is an AI, that the interview is recorded and scored, how many questions there are and that some are timed. Several vendors put the consent on the start screen.
2. 2The screening questions. Work authorization, location, start date and sometimes pay, as one-line facts.
3. 3Questions built from the posting. Usually three to eight, each tied to a requirement or a line of your resume, with one or two follow-ups.
4. 4A scenario or a how-would-you question. Something the role would hand you, and what you would do first.
5. 5The close. Anything to add, then the end. No result on the call.

The posting questions carry the most weight. HireVue says its AI suggests questions and answer rubrics from the job description, which the employer approves. A typical one reads “The role requires SQL. Tell me about your experience with it.” The follow-ups are plain: “What did you do first?” is Ribbon’s own example, and “What was your role in that specifically?” and “What was the result?” are the ones candidates report most.

Then there is the clock. At micro1, a custom question has a limit the employer sets, from one to four minutes, and at the limit Zara stops the response and moves on, according to micro1’s help pages as quoted by labelingjobs. A candidate on Blind put it more bluntly: it would cut you off while you were answering. Silence has rules too. Mercor’s candidate FAQ warns that extended silences may trigger an interjection, and says you can ask it to repeat or clarify a question.

The whole round is short. Ribbon says its interviews typically take around 20 minutes, and Mercor says most take about 20. Across 172,486 interviews on the Flowmingo platform, the median was 25 minutes from start to submit, setup included. Questions about the role, the team or the process are usually deflected to the hiring team. Nobody gives you a result on the call.

The scoring the vendors describe reads the transcript. Each requirement gets a rating against a rubric: HireVue offers Meets or Does Not Meet, or a custom 1 to 5 scale, and Ribbon gives recruiters a recording, a transcript, a summary and 1 to 5 scores. Alex produces a transcript and a score for every interview and says it does not make final hiring decisions. Ribbon says the same of its scores: the decision stays with the recruiter.

Alongside the scores come flags. Ribbon’s integrity monitor flags off-screen activity and answers that mimic AI-generated structures. Mercor says it flags submissions that are evidently produced by AI. Some vendors also check your answers against your resume, so a date that does not match is worth fixing before the interview, not during it.

HireVue’s FAQ says it never uses the video recording to identify you, and none of the vendor pages we read describes scoring your face or your accent. What they describe is matching: what you said against what the requirement asked for. That is why the preparation below is mostly about saying the match out loud.

What happens next is often nothing you can see. In Greenhouse’s survey, of the candidates who completed an AI interview, 28% moved forward, 13% were rejected and 51% never heard back.

The rules depend on where you live, and this is not legal advice. In Illinois, the Artificial Intelligence Video Interview Act has required since 2020 that an employer using AI to analyze a video interview tells you, explains what the AI evaluates and gets your consent first. New York City’s Local Law 144 requires a bias audit of automated hiring tools and, in most summaries, notice to candidates before one is used. In the EU, the AI Act treats AI used in hiring as high risk, with transparency duties from August 2026. Other states have bills pending.

Some vendors offer a way out on their own. HireVue says candidates can opt out, with an equivalent path that does not use AI, and that accommodations are available. In Greenhouse’s survey, 46% of candidates wanted the option of a human interview. Asking the recruiter whether one exists, and whether a person reviews the AI’s results, is a reasonable question.

The guides we read agree on this list. Most of it is about making your answer easy to match, because that is what the scoring does.

1. Give the direct answer first. A person can infer that your story answers the question. The scoring compares your words with the requirement, so say the match in your first sentence, then give the example.
2. Use the posting’s words. If the posting says “stakeholder management”, say “stakeholder management” before you describe the meeting. The rubric was written from that posting.
3. Say “I”. Use “we” for the team’s goal and “I” for what you did. A transcript full of “we” gives the screener nothing to credit to you.
4. Put a number on the result. Hours saved, tickets closed, a percentage. If there is no number, name what changed and who noticed.
5. Keep answers to 60 to 120 seconds. That is the advice in every guide we read, and it fits most limits. Practice with a timer, because the first try almost always runs long.
6. Say “let me think for a second” instead of going quiet. A long silence can be read as the end of your answer, or prompt an interjection.
7. Ask for a repeat rather than guess. Mercor and micro1 say their interviewers repeat or clarify a question when asked. An answer to the wrong question scores as a miss.
8. Camera on, face in frame, one screen. Close other windows, unplug the second monitor and sit where your face stays visible. Off-screen activity is one of the things vendors flag.
9. Take the vendor’s practice run. Mercor offers unlimited practice interviews and micro1 a practice interview of its own. They show you the setup, the voice and the timer before it counts.
10. Use no AI help during the interview. The vendors prohibit it, and several say they detect and flag it. Prepare before; answer on your own.

Practice cannot give you the employer’s questions. They come from a rubric you will not see, and HireVue’s own FAQ says no course will help you game its assessment. What practice can do is make the habits above automatic: the direct answer first, “I” instead of “we”, a number, and an answer that ends before the clock does.

Star Interview’s AI screening interview is a 20-minute practice run of this round by voice. An AI interviewer, which says it is one, asks the screening questions, your background, three questions built from your job description, a scenario and pay, with each timed answer on the clock and a cut at the limit. Afterwards you get a report on what a screener would have matched, missed and flagged, and how much of each limit you used. There is a 30-minute version with two questions on how you would do the work, and a five-minute drill of one timed answer.

Two limits are worth knowing. The real round is often on video with your camera on; this practice is by voice, so check your camera and your screen separately. And Star Interview is not affiliated with any employer’s interview vendor. Practice before the real one, never help during it.