What to expect in an AI interview: retakes, scores, humans and fairness

An AI interview follows the same outline almost everywhere. You get an invitation with a link. You grant camera or microphone access in the browser. You answer questions, spoken or typed, with nobody from the company on the other end. The platform turns your answers into a transcript and a report, and a person at the employer reads it and decides what happens next.

What differs is the shape. The table below is taken from each platform's own pages, read on 28 September 2026. Career1 is not affiliated with any of the others.

Swipe the table sideways to see every column.

For the questions themselves, and the difference between a recorded interview and a conversation, read AI interview questions. The rest of this page covers what that article does not: the rules, the humans and the worries.

Retakes are set by the platform and often by the employer. Mercor says "Candidates are entitled to retake an interview up to three times across all applications related to that particular interview." HireVue's 2025 AI Explainability Statement says its AI-scored interviews "allow candidates multiple attempts to answer each question", and lists unlimited retries as its recommended employer setting. Spark Hire says you will have "anywhere from 1 to unlimited attempts" depending on the company. micro1 does not publish a retake policy on the pages we read. Career1's vetting interview is one attempt, with no retakes; practice is where retries live.

Fumbling one answer matters less than it feels in the moment. A conversational interview is a transcript of many answers, and one weak one is one part of it. What helps:

• Correct yourself out loud. "Let me put that better" is fine, and the correction is in the transcript too.
• Ask for the question again. Mercor's help page suggests you "Request the AI to repeat or clarify a question if necessary". Career1's interviewer offers to repeat a question when you go quiet.
• Keep talking if it cuts in. Mercor says "If the interviewer interrupts due to a pause, you can let it know you weren't finished." Career1's interviewer is told to stop and listen when you interrupt it.
• Do not end it early to start over. Career1's end button warns that you cannot restart the interview afterwards.

If you struggle on a Career1 question, the interviewer is told to simplify it rather than move on. For answer technique, read how to pass an AI interview.

During the interview, yes, you are talking to software, and it should say so. Career1's interview room says it is an AI before you start, and the interviewer confirms it if you ask. Afterwards, on every platform we checked, a person is involved somewhere.

• HireVue's candidate FAQ says "those in the top group move on to person-to-person interviews before hiring decisions are made by people."
• micro1 says "the session is recorded so our human recruiters can review results and match candidates to roles", in its candidate documentation. Its candidate privacy policy adds that "The overall assessment will be reviewed by a human evaluator before a final decision is made."
• Mercor's data page says human contractors sometimes help assess interview performance, and that their feedback "is never utilized for making hiring decisions".
• Career1: for a job application, the hiring company's team reads the report with the transcript and recording beside it, and a person there makes every decision. Nothing rejects anyone automatically. Career1's own staff do not review or grade individual reports, but if you think your vetting report misstates what you said, a person at Career1 will compare it with the recording and correct it.

"A human sees it" does not always mean a human watches all of it. On a ranked list, people read from the top. That is why the score matters, which is the next question, and why replies can take weeks, which heard nothing after an AI interview? deals with.

It means different things on different platforms, and some do not show you one.

On Career1's vetting interview, the report gives an overall score, a technical score and a communication score, each out of 100, plus each skill on your resume marked verified or not, with the quote from your interview behind it. There is no pass mark: finishing the interview verifies your profile whatever the score. But be clear about what the score does. Companies searching the pool see it, profiles are ranked by relevance with the score deciding ties, and a company can choose to see only profiles above a score it sets. You can read your whole report on your dashboard. The full breakdown is in AI interview questions.

On a Career1 job application, the company sees a match to the role, an overall score and a recommendation. You see your application's stage, not the score.

Elsewhere: HireVue's explainability statement recommends that employers show result tiers, labelled for example top, middle and bottom, and lists numeric scores as "Recommended Off". micro1 gives "A real-time result indicating whether you meet micro1's global certification criteria", according to its interview guide, without publishing the criteria. Mercor does not publish a pass mark.

So a score is an ordering signal for a person, not a verdict. It is also not a measure of your worth: it is one short conversation, read by a model.

This deserves a straight answer, including the part that is not reassuring.

What Career1 evaluates. The words of the conversation and your resume. Nothing in Career1 scores your face, expressions, eye contact, tone of voice, accent, background or camera, and pauses do not appear in a transcript. Communication is judged from how clearly your answers read, so an answer that loses its thread can score lower; nerves themselves are not measured.

Where it can still affect you. Your speech is turned into a transcript by a speech-to-text model, and the report is written from that transcript. If a word is misheard, the report reads the misheard word. The interviewer also decides your turn has ended after a silence of under a second, so a long block in a sentence may bring in the next question. If it does, keep talking: it is told to stop and listen. After a vetting interview you can read the transcript on your dashboard, and ask a person to check the report against the recording if something was transcribed wrongly.

Ask before you start, not after. The spoken format does not suit everyone. Career1 shows each question as text once the interviewer has said it, but it does not caption your own speech and has no written alternative today. If you need an adjustment, email hello@career1.ai before you use your one attempt. Employers using Career1 are responsible under its terms for reasonable adjustments, such as for a disabled candidate who cannot take a spoken interview.

Other platforms, in their words. HireVue's explainability statement says its AI "does not use any video analysis or other audio characteristics", including "tone-of-voice", and that it tested transcription for non-native English speakers "with a variety of accents". Its FAQ tells autistic candidates they "may also be able to ask for the interview questions in advance" and can "request accommodations". Sapia.ai is typed and says it "is not timed". Mercor turns captions on by default.

It depends on the platform, and on settings the employer chooses.

• Career1 does not track tab switches, watch your screen or count windows. Refreshing reconnects you to the same interview rather than starting a new one, and what you already said stays in the transcript. The browser asks before you leave the page. A vetting interview can reconnect up to five times; after that it pauses, and you email hello@career1.ai to have it reset. A practice interview can be rejoined within 24 hours of starting it.
• HireVue offers employers "browser-focus tracking" and "automated flags for browser-focus loss or unexpected session changes", according to its cheating and fraud page, which also says "A single behavior should not automatically be treated as proof of fraud." If your connection drops, its FAQ says you can "pick up where you left off with the same link".
• micro1 lists data from "proctoring tools to assess the integrity of your interviews", without saying which signals they use.
• Mercor says that if your tab freezes, you reopen your Assessments tab and try again, or retake if the session has ended.

The safe routine is the same everywhere: close other tabs and notifications before you start, do not refresh unless the interview is stuck, and if something breaks, take a screenshot and email the recruiter the same day.

You can ask. Whether you get one depends on the employer and the platform.

• HireVue's explainability statement says candidates can opt out of AI evaluation, in which case "recruiters manually review and rate these candidates' interview responses" against the same scales.
• micro1's candidate privacy policy says that if you are not selected and want to appeal its use of automated technology, you can email privacy@micro1.ai.
• Mercor says "Our AI interview is the standard and only method for screening candidates" and that "interviews cannot be conducted via chat, call, or any other medium."
• On Career1, a company's job belongs to that company, so ask the recruiter. The vetting interview is optional: nobody has to take it, and there is no human version, but you can email before your attempt if you need an adjustment.

A short message works: "I would find a spoken AI interview difficult because of a disability. Is there an alternative format, or an adjustment such as extra time or the questions in advance?" You do not have to give a diagnosis to ask.

If the role is real and you want it, usually yes. On platforms such as Mercor, where the AI interview is the only screening method, declining ends the application. Before you start, a few checks take five minutes.

1. Find the job on the company's own careers page. A real vacancy is listed there, or the recruiter can point you to it.
2. Check who is inviting you. The email should come from the company or from the platform it names.
3. Know what is recorded. The platform's candidate page should say. Our guide to what happens to your AI interview recording sets out five platforms' terms side by side.
4. Never pay. Candidates do not pay for a real interview. Career1 is free for job seekers, always.
5. Know the rules on AI help. They differ by platform, as can you use ChatGPT in an AI interview shows.

Then practise once, out loud. Career1's free AI mock interview is private, once a week, with no video and no score. When you are ready for the interview that counts, getting vetted explains the one-attempt vetting interview, and the FAQ answers the rest. Career1 is new: no company is hiring through it yet, so being in the pool means companies can find you, not that a call is coming.

Every source below was opened and re-read on 28 September 2026.

• Mercor, AI Interview help page and How Mercor Uses AI and Data
• micro1, library introduction, AI interview guide and Candidate Privacy Policy, last updated 21 July 2026
• HireVue, candidate FAQ, cheating and fraud mitigation and 2025 AI Explainability Statement
• Sapia.ai, Candidate Explainer
• Spark Hire, How to ace a one-way video interview