How to use AI for interview prep

Give the AI the job, not a job

Paste the whole posting and your resume before asking anything. Generic AI interview prep produces generic questions; the value is in the overlap between what the role wants and what your history shows. Ask it to find the gaps first.

Here is a job posting and my resume. List the five requirements in the posting that my resume shows weakest evidence for, and for each one say which line of my resume an interviewer would probe.

[paste posting]

[paste resume]

Have it write the questions this interview will actually ask

Not "common interview questions". Ask for the ten questions a hiring manager for this role at this company would ask, with the reason behind each. Half of them will be behavioral; keep those for step four.

Act as the hiring manager for this role. Write the ten questions you would ask in a 45-minute first-round interview, in the order you would ask them. After each question, in one line, say what a weak answer and a strong answer would sound like.

Build six stories, not thirty answers

Behavioral questions are answered from a small bank of stories. Have the AI read your resume and propose six, each tagged with the two or three competencies it can cover. Then correct it: it only knows what is written down, and the best stories usually are not.

From my resume, propose six stories I could use in behavioral interview questions. For each: a one-line title, the competencies it demonstrates (ownership, failure, conflict, leadership, pressure, teamwork), and the one number I should be able to quote from it.

Answer out loud, not in the text box

This is the step most people skip and the one that matters. Typing an answer produces a written answer. Interviews are spoken. Set a two-minute timer, say the answer to the wall, and only then compare it with what you would have written. The gap is what you are fixing.

Where a chatbot stops: A text chatbot cannot hear you. Voice mode helps, but it still waits for you to finish and never interrupts. That is the next step.

Make it push back

Real interviewers follow up. "Tell me about a project you led" is always followed by "how many people?", "what went wrong?", "what was the result, in numbers?". Tell the AI to do the same and to refuse vague answers. Then notice how quickly a chatbot forgets the instruction.

Interview me for this role, one question at a time. After every answer I give, ask one follow-up that demands a specific: a number, a name, a date, or what went wrong. Do not move on until I give one. Do not praise me. Start.

Where a chatbot stops: Chatbots drift back to being agreeable within a few turns. A dedicated AI interviewer with the pushback built into the character does not, and is the reason our interviewer, Dana, exists.

Get graded against a standard, not encouraged

Ask for a score and a reason, and forbid compliments. Better still, ask what the interview would have produced: no follow-up, a second round, or an offer. That framing forces the AI to judge the outcome rather than the effort.

Grade the interview so far. Give a score out of ten, then the three most important reasons for that score, most important first. No encouragement. Finally, say which of these you think would have happened after this interview: no follow-up, kept on file, second interview, offer.

Change one thing and run it again

One run tells you where you are. Two runs a week apart tell you whether you moved. Fix the single biggest reason from the last grade, leave everything else alone, and repeat. Keep the transcripts; the difference between run one and run four is the most convincing thing you will read all month.