How an AI interviewer actually scores your answer

Candidates ask us the same question in different words: is the score fair, and what is it based on? Hiring managers ask the mirror image: can I trust it, and can I check it? Both deserve a concrete answer, so here is how a SealHire assessment is built — from the brief the interviewer starts with to the report both of you read.

It starts with the brief, not with you

Before the interview begins, there is a role brief: what the role involves, what a strong candidate shows, and a short list of skills the interview will assess — typically five. For a mock interview the brief comes from the role catalog and you pick the level. For a real interview the employer set the brief from their job description and chose the level they are hiring at.

This matters because the assessment is not "how good is this person" in the abstract. It is "for this role, at this level, what did the candidate show on each of these skills?" A brilliant answer about something the brief does not care about earns a note, not a score.

The interview is a search for evidence

Alex, the interviewer, opens each skill area with a question about the work — how you do something, a time you had to, what usually goes wrong. Then it follows up on what you actually said. The follow-ups are not random. They go where the evidence is thin: a claim without a detail, a result without a cause, a decision without a reason.

The reason is simple. The assessment can only score what was said. "I improved the onboarding flow" contains no evidence about any skill. "The onboarding flow lost most new users on the second screen, I rewrote it as three short steps with a progress bar, and completion went up within the first week after we shipped it" contains evidence about product judgment, execution, and measurement — three things a brief might well be asking about.

So the interviewer probes. If you answer the follow-up with detail, the detail becomes evidence. If you generalise again, the skill is scored on what is there, which may be very little. The single most useful thing to know about AI interview scoring is this: specifics are the currency.

A human interviewer forms an impression. The assessment collects evidence. Everything about how to do well follows from that one difference.

What a verdict looks like

At the end, the assessment produces one verdict per skill in the brief. Each verdict has three parts:

1. The judgment — how well the skill was evidenced against the level in the brief.
2. The evidence — the quotes from your transcript that support the judgment, each linked to the moment in the recording. A reader can click the quote and hear you say it.
3. The reasoning — why those quotes led to that judgment, in plain language, including what was missing.

The third part is what most candidates find useful. It says things like "the candidate described the outcome clearly but did not explain the decision that produced it" — which is exactly what to fix before the next attempt.

Above the skills there is an overall recommendation for the hiring team, and a section for integrity signals: whether it looked like answers were being read, whether tabs were switched, whether someone else appeared to be answering. Those are observations with timestamps, not accusations, and a human decides what they mean. They exist so the hiring manager can trust that the evidence in the report was produced by you, live.

A single skill, scored two ways

Take one skill from a project manager brief — stakeholder communication — and one opening question: "Tell me about a time you had to deliver bad news to a stakeholder."

Candidate one says: "I believe in being transparent, so I always tell stakeholders early when something is going wrong. Communication is one of my strengths." The interviewer follows up: "Can you give me a specific time?" The candidate: "It happens fairly often, I just make sure people know." The verdict on this skill will say the candidate expressed a principle but gave no instance, so the skill could not be evidenced at the level — and the reasoning will say exactly that, so the candidate knows what was missing.

Candidate two says: "Last year the data warehouse migration was going to miss the quarter-end reporting deadline. I found out on a Tuesday, and I put a fifteen-minute call in the finance director's diary for that afternoon rather than waiting for the Friday status meeting. I brought the two options — a partial cutover with manual reconciliation, or a full slip of two weeks — and a recommendation. She chose the partial cutover, and we agreed the reconciliation would be my team's job, not hers." The verdict quotes the Tuesday call, the two options and the recommendation, links each to the recording, and the reasoning explains why that is evidence of the skill at a senior level: speed, options rather than problems, ownership of the consequence.

Neither candidate is smarter than the other. One of them gave the assessment something to score.

Why nothing in the report can be edited

The moment the interview ends, the recording, the transcript and the assessment are sealed. Nobody can change them — not the employer, not SealHire, not you. If you retake a mock interview, the new attempt is a new sealed record; the old one is untouched.

This is what makes the report worth reading. A résumé is a claim; a sealed interview is a record. When you share a mock interview with a hiring manager, they are not reading a document about you that you wrote. They are watching you answer questions you did not see in advance, with an independent assessment attached that you could not have altered. That is a stronger thing to send than any summary of yourself, and it is why SealHire's candidate side is built around sending interviews instead of résumés.

What the score is not

A few things the assessment deliberately does not do:

• It does not score fluency. Pace, accent, filler words and thinking pauses are not part of any verdict. The transcript is about what you said.
• It does not score confidence. There is no column for it. A confident answer with no evidence scores as an answer with no evidence.
• It does not compare you to other candidates. Each report is against the brief. Two candidates for the same role are compared by the hiring team reading two reports, not by the assessment ranking them.
• It does not make the hiring decision. The report ends in a recommendation with the evidence behind it, and a person reads it and decides. Every part of the report is inspectable precisely so that the person can disagree with it.

How to use this if you are the candidate

Read the brief. For each skill, prepare one real example with details. In the interview, answer the question asked, give the detail, end on the result, and say what you would change. Then read the report the way a hiring manager will: skill by skill, quote by quote. Where the reasoning says evidence was missing, you now know exactly what to add.

Then take it again. The first attempt teaches you the format; the second is the one you keep. Your first three mock interviews are free, and one is only used when you finish it.

How to use this if you are the hiring manager

Do not read the number first. Read the evidence for the two skills you care most about, click through to the moments in the recording, and decide whether you agree with the reasoning. The report is built to be argued with — that is what "evidence-backed" means. If you find yourself agreeing with the reasoning across a few candidates, you can start trusting the summary. If you do not, tell us; the reasoning is where we improve the interviewer.

Either way, the thing you are looking at is a record of a real conversation, and every claim in it can be checked against the moment it was made. That is the whole point.