Eleven mistakes candidates make in AI interviews — and what to do instead

We read a lot of interview transcripts. Not to judge the people in them — the assessment does that, skill by skill, with the evidence attached — but to make the interviewer better. Along the way a short list of avoidable mistakes keeps appearing. None of them is about talent. All of them are about how a spoken, structured interview differs from the interviews people are used to. Here they are, with the fix for each.

Before the interview

1. Treating it like a one-way video

A one-way video interview asks a fixed question and gives you a fixed time to record an answer. An AI interview on SealHire is a conversation: the interviewer, Alex, listens to your answer and builds the next question from it. Candidates who prepare for the first format walk in with rehearsed monologues and then get thrown when the second question is about a detail in their own first answer.

Instead: prepare stories, not speeches. Know the situation, what you did, what happened, and what you would change — and be ready to go one level deeper on any of it. Our preparation checklist walks through how.

2. Not reading the role brief

Every mock interview on SealHire is built from a role brief you can read before you start: the work, what a strong candidate shows, the skills that get scored. Candidates skip it and then spend the interview discovering what it is about. The brief for a Registered Nurse is not the brief for a Sales Development Representative, and the questions follow the brief.

Instead: open the role page, read the five skills, and decide which real example proves each one. Five lines on a piece of paper is enough.

3. Picking the wrong level

You choose the seniority level before a mock interview starts, and the questions and the bar move with it. Picking "senior" because it sounds better, when you are interviewing for mid-level roles, gets you a report that says you are not yet senior — which is true, and not what you needed to learn.

Instead: pick the level of the jobs you are actually applying for. If you are between two, take the interview at both; the first three are free.

During the interview

4. Answering the question you wish you had been asked

The most common pattern in transcripts that score poorly: the interviewer asks about a specific thing — a decision, a failure, a trade-off — and the candidate answers with a general description of their experience. The follow-up asks again, more specifically. The candidate generalises again. The skill ends up scored on no evidence, because none was given.

Instead: answer the question that was asked, with a specific example, then stop. If you do not have an example, say so and offer the closest thing you do have. An honest "I have not done that exactly, but here is the nearest situation" is scored on the nearest situation.

The interviewer is not trying to catch you out. It is trying to find the evidence. Give it something to find.

5. Reading from a script

The interviewer notices reading. The rhythm changes, the eyes drop, and the answer arrives in complete paragraphs that do not match the question. SealHire notes it, with a timestamp, in the integrity section of the report that the hiring manager reads. It does not decide anything — a human does — but "answers appeared to be read aloud" is not the note you want next to your name.

Instead: a one-line prompt sheet is fine. Glance at it to remember which story you planned. Then look up and tell the story.

6. Switching tabs

Same section of the report. Tab switches and window changes are observed and noted. Most of the time the candidate was checking a fact, and most of the time it did not help the answer.

Instead: close everything else before the technical check. If you do not know a fact, say "I would check that" — it is a better answer than a suspicious pause.

7. Trailing off instead of finishing

Spoken answers end in one of two ways: a clear finish, or a slow fade — "…so, yeah." The interviewer waits through pauses, so a fade is not cut off, but it leaves the answer without a conclusion and leaves the interviewer to guess whether you were done.

Instead: end answers on the result. "It hit the deadline" or "we did not ship it, and here is what I took from that" tells the interviewer you have finished and gives the assessment a clean landing.

8. Filling every silence

Some candidates hear a pause and rush to fill it, adding a second example, then a third, then a caveat about the first. The interviewer gives you room to think; you do not have to use all of it talking.

Instead: it is fine to say "let me think for a second." Thinking pauses are not scored. Rambling is not scored either, but it uses up the minutes that another question could have used.

9. Never saying what you would do differently

The interviewer asks some version of "what would you change?" about almost every example. Candidates who answer "nothing, it went well" have missed the point of the question, which is about judgment, not about the outcome.

Instead: have the reflection ready before the question comes. One sentence — "in hindsight I would have involved the client earlier" — reads as maturity, and it is usually true.

10. Performing confidence instead of showing evidence

An AI interviewer does not respond to charm the way a person might, and the report does not have a column for it. What the assessment can see is the evidence: what you said you did, how specific it was, whether it held up under a follow-up. A confident non-answer scores as a non-answer.

Instead: spend your energy on details — names of systems, sizes of teams, real constraints, actual numbers you know. Detail is what confidence looks like in a transcript.

After the interview

11. Treating the first attempt as the final one

A mock interview is practice. The report tells you which skills were evidenced and which were not, with the quotes to prove it. Candidates who read it, shrug, and share it anyway are handing a hiring manager their draft. The candidates who get the most out of the format read the report, prepare the two weak skills, and retake it.

Instead: take it once to learn the format, once to get it right, and keep the one you are proud of. When you share it, the hiring manager sees a sealed interview and an assessment — not the attempts before it.

The same mistake, before and after

Here is mistake number four as it actually sounds. The question: "Tell me about a time a project you owned slipped. What happened?"

Before: "I've managed a lot of projects and generally they've gone well. I'm quite organised, I use a tracker, and I keep stakeholders informed, so slips are rare. When they happen I communicate early." Nothing in that answer can be scored — there is no project, no slip, no decision.

After: "The billing migration last spring. We planned six weeks and it took nine. The slip came from a dependency we did not own — the payments vendor's sandbox was down for most of week three — and I made the call to keep the team on the parts we could test rather than pause. I told the stakeholders in week three that we would miss by two weeks, and we missed by three. What I would change: I would have asked the vendor for their sandbox status history before we planned around it." Same person, same experience, now with a project, a cause, a decision, a result and a reflection — five things the interviewer can follow up on and the assessment can quote.

The pattern underneath all eleven

Every mistake on this list is a version of the same thing: treating the interview as a performance to get through rather than a chance to put evidence on the record. The interviewer is patient, it is not trying to catch you out, and the report it produces is the most detailed feedback most candidates will ever get on how they interview. Use it that way.

If you have not taken one yet, pick your role and start. Thirty minutes, free, and nobody sees it until you decide they should.