AI Interview Practice: A Mock That Talks Back

Most people prepare for an interview the same way. They find a list of the 50 most common questions, write an answer under each one, and read the answers until they feel familiar. Then the interviewer asks the first question, nods, and says "can you give me a specific example?" and the prepared answer has nothing left to give.

That is the gap AI interview practice can close, if you use it properly. A list of questions cannot ask you a follow-up. A mock interviewer can. The value is not in hearing the question. You already know the questions. The value is in being made to answer the second one, and the third.

This post covers what a useful mock interview looks like, the follow-up questions that catch people out, and the mistakes that show up again and again when you actually run one.

Why question lists stop working at question two

Interviewers do not grade your first answer as much as you think. The first answer is a starting point. What they are really listening for is what happens when they dig.

A typical exchange goes like this (illustrative):

Interviewer: Tell me about a project you're proud of.

Candidate: We rebuilt the reporting dashboard for the operations team, and it made things a lot more efficient.

Interviewer: What was your part in it?

Candidate: I was involved in most of it. We worked really closely as a team.

Interviewer: And how did you know it made things more efficient?

By the third question the candidate is improvising, and it shows. Nothing they said was false. It just was not specific, and every follow-up made that more obvious.

If you only rehearse first answers, you are rehearsing the easy part.

The four follow-ups to prepare for

Almost every follow-up an interviewer asks is one of four kinds. If you can answer these for each of your main stories, you are ready for most of what comes.

1. "What did you do, specifically?"

This is the "we" problem. Teams do work, but interviewers hire individuals. For every story, have one sentence that starts with "I" and names an action only you took.

I wrote the queries behind the three main charts and ran the sessions with the shift leads to decide what they actually needed to see.

2. "How did you measure that?"

"More efficient", "better", "faster" all invite this question. Attach a number or an observable change, even a small one.

The weekly report used to take about half a day to put together by hand. After the dashboard, it took nobody any time, because it updated itself.

If you do not have a number, describe what changed in a way someone could check. "The shift leads stopped asking us for the spreadsheet" is evidence.

3. "What would you do differently?"

This is not a trap. It checks whether you learn. Have a real answer that is about your own judgement, not about a colleague or bad luck.

I'd talk to the shift leads earlier. We built two charts nobody used because I assumed what they wanted.

4. "Why does that matter for this role?"

The question behind every question. Connect the story to the job in one sentence, using a word from the job description if you can.

This role is mostly about turning messy operational data into something people decide with, which is exactly what that project was.

A mock interview script you can rehearse

You can use this structure with a friend, with a general AI chat, or with a dedicated tool. The structure matters more than the tool.

Set the scene. Give the interviewer the role, the company type and your background in two or three sentences. A mock without context asks generic questions, and generic questions get generic answers.

Ask for follow-ups, every time. The single most useful instruction is: "after each answer, ask at least one follow-up that digs into what I just said." Without it, most mock interviews become a quiz.

Answer in full sentences, at real length. Type or say what you would actually say in the room. If you write notes instead of answers, you are practising notes.

Do not restart when it goes badly. The awkward moment is the training. Answer the follow-up you were dreading, even badly, then fix it next round.

Get feedback on the whole exchange. Not just "good answer". Ask which follow-up you handled worst and what a stronger answer would have contained.

Then run it again.

Mistakes that come up in almost every mock

These are the patterns worth catching before the real interview.

• Reciting your CV in order. "Tell me about yourself" is not a request for your history from university onwards. Lead with what you do now, one result, and why you want this job. Keep it under two minutes.
• Saying "we" all the way through. Covered above. If the interviewer has to ask what you did, you have already lost a little ground.
• Achievements without evidence. "I improved the process" with no number, no before-and-after and no example.
• The non-weakness. "I'm a perfectionist" or "I work too hard" signals that you prepared a line rather than thought about the question. Name a real, minor weakness and what you are doing about it.
• No questions at the end. "No, I think you covered everything" ends the interview on a flat note. Have two questions ready, one about the work and one about how success is judged in the first months.
• Dodging the follow-up. When you do not have a good answer, changing the subject is more noticeable than admitting it and giving the best honest answer you have.

AI interview practice with Codeset

A pushback you have heard before is a pushback you can handle. That is the same idea behind rehearsing how to ask for a raise, where the conversation is decided by your answer to the first "no", and it is why a phrase list for ordering food in Spanish only gets you so far once the waiter asks something unexpected.

This is what we built Codeset for. You describe the goal in your own words. We typed "Practise a mock job interview for a product analyst role and handle tough follow-up questions". The tutor asked one question about our starting point (three years as a data analyst at a logistics company, first product analyst interview next week, nervous about "tell me about yourself" and the follow-ups), then planned an 18-lesson route of about ten minutes each.

The route starts with the follow-ups, not the opening answer: "Answering Metric Follow-Ups" is lesson 1, and "Structuring Tell Me About Yourself" comes fourth.

Each lesson ends in a typed role-play with an interviewer who has a role and an agenda. Here it was Jordan, a product manager who "probes the candidate's metric choice and asks how their logistics experience applies". The opening question was a fair one: the company is considering a clearer delivery-tracking page, so what metric would tell you it is working?

We played the candidate who makes most of the mistakes in the list above. Page views as the metric. "A better customer experience" as the outcome. "We" for everything, no numbers, a perfectionist weakness, and no idea what the metric might miss.

Jordan keeps narrowing the question. When the candidate says page views, he points out that customers might open the page repeatedly because they still cannot tell where their delivery is.

What is worth noticing is how the interviewer behaved. Jordan did not move on to the next question on a list. He followed up on each weak answer, and when the candidate switched to "time spent on the page", he answered: "More time on the page could mean customers are struggling." By the fifth turn he was close to handing over the answer, suggesting delivery-related support contacts as a measure. The candidate agreed, said "We would work it out as a team", and ended the interview. That is how real interviews go wrong: not with a trick question, but with a reasonable follow-up you have no answer for.

15/100: "You showed limited progress toward a useful metric, but did not clearly define the customer outcome, explain a trade-off, or connect your logistics work to a specific analysis."

The tutor scored it 15/100. "Explains a limitation" got 0, which is follow-up 3 above in action: the candidate could not say what their own metric would miss. The three fixes it suggested are specific enough to rehearse straight away:

• "State that the page should help customers find a reliable delivery status without contacting support."
• "Use the share of tracking visits followed by no delivery-related support contact within 24 hours, and explain what that measure could miss."
• "Give a specific example of analyzing delivery outcomes alongside support patterns, while being clear about your role."

The last one is the "we" problem again, caught by the tutor without being asked. Then you press "Try again" and answer Jordan properly.

Practice is typed for now, not spoken. That still does the main job, because the hard part of a follow-up is knowing what you would say, and you find that out by being asked.

FAQ

Is AI interview practice as good as practising with a person?

It is different. A person who knows your field, or who has hired for the role, can tell you things an AI cannot. An AI mock is available whenever you are, does not get tired of asking the same follow-up, and does not go easy on you to be polite. Most people benefit from both, and the AI one is easier to do five times.

Should I memorise my answers?

Memorise your stories, not your sentences. Know the situation, what you did, the result and the lesson. If you memorise word-for-word, the first unexpected follow-up knocks the whole script over.

Rehearse the second question

You already know the first question in your next interview. What decides it is the one after. If you want to practise those with an interviewer who keeps digging, try Codeset. Every account starts with 1,500 free credits and no card.