Preparing for an Interview With AI

The common approach is to ask a chatbot "how do I answer "tell me about a weakness"", get a polished paragraph, and memorise it. It fails for a specific reason: the answer contains no evidence from your life. Interviewers are not really evaluating the answer, they are evaluating what happens when they ask the follow-up. A memorised paragraph has nothing underneath it, and the second question exposes that immediately.

Used well, AI is not an answer generator. It is a way to interrogate your own material faster than you could alone — finding the examples you have forgotten, turning them into something coherent, and stress-testing them before someone else does.

Before you write a single answer, gather the raw material. This is the step people skip and it is the one that makes everything afterwards work.

Your resume, obviously — but also the longer version you trimmed down.

Performance reviews and any written feedback you have received.

Project documentation, design docs, post-mortems, retrospectives.

Actual numbers: latency before and after, revenue, headcount, timelines, error rates, user counts.

For engineers, the repositories you worked in. For everyone else, the decks and spreadsheets you produced.

Upload all of it into a tool that can search it. JobSeeker AI indexes PDF, DOCX, PPTX, TXT, CSV, HTML, Markdown, JSON and SQL, plus source files across more than 40 programming languages, and keeps the index on your own machine. The point is that when you later ask "what did I actually do that shows ownership", the answer comes from your documents rather than from a model's imagination.

Do not prepare for interviews in general. Prepare for this one. Paste the job description in and ask what a competent interviewer would probe given that specification and your background.

Ask for three categories separately, because they fail in different ways:

Behavioural — ownership, conflict, failure, influence without authority, competing priorities.

Technical or domain — drawn from the specific tools and responsibilities the posting names.

Gaps — the uncomfortable ones. A requirement you do not meet, a short tenure, a career break, a title that does not match the level. These are the questions people avoid preparing and then fumble.

Twenty to thirty questions is a realistic target. You will find the same four or five underlying stories answer most of them.

Now work question by question, and phrase the request so the answer is forced to come from your material: "using my project documents, what is my strongest example of resolving a technical disagreement?"

A usable structure for behavioural answers:

The "why that rather than the alternative" clause is what separates a real answer from a recital. It demonstrates judgement, which is the thing actually being assessed.

This is the highest-value step and the one nobody does. Take each drafted answer and ask the AI to act as a sceptical interviewer: what is the weakest claim here, what would you probe, where does this sound rehearsed, what evidence is missing?

You will find the same problems repeatedly — credit taken for team work, a result with no baseline, a decision presented as obvious when it was actually contested. Fix them now, at your desk, rather than discovering them live.

Then ask for the three follow-up questions each answer invites, and prepare those too. Interviews are won in the follow-ups.

An answer that reads well on a screen frequently collapses when spoken. It runs ninety seconds too long, the middle wanders, you say "basically" eleven times.

Say each answer aloud and transcribe yourself. Reading your own spoken words is uncomfortable and extremely effective — the padding, the circling back, the point that never quite arrives all become obvious. JobSeeker AI's live transcription works fine for this: turn it on, answer a question to an empty room, and read what you actually said.

Target sixty to ninety seconds for behavioural answers. Rehearse until the structure is automatic and the words are not. You want to know the shape of the story, not a script.

"Do you have any questions for us" is an evaluated part of the interview, and "no, I think you covered everything" is a bad answer. Use what you have already loaded: given this job description and this company, what would a genuinely interested candidate want to know that is not answered publicly?

Take five in, expect to ask two. Good ones are specific — how the team decides what to build, what happened to the last person in the role, what the first ninety days look like, what makes this job harder than it appears.

Roughly six hours spread over a week. The final point matters: cramming on the morning makes you sound rehearsed and anxious. The work is already done by then.