How Close Is a Simulated Interview to the Real Thing? Honestly.

The interview I couldn't finish

There's a kind of interview that bothers me more than a bad one. I can feel that the person across from me is a good engineer. And I still can't check it.

Not because they're hiding anything. Because they didn't prepare, and every answer comes out as a shape with nothing inside it.

When that happens I start helping. I narrow the question. I hand them the opening they need, the one where all they have to do is describe something they actually did. Sometimes it works. Often it doesn't, and we get to the end of the interview having produced nothing either of us can use. They leave without an offer, I leave without an opinion.

An unprepared candidate doesn't just interview badly. They make the interview unverifiable. Nobody learns anything about them, including the person who was trying to.

It's worse in IT than almost anywhere else, and I think that's because engineers treat interviewing as an interruption rather than a skill. Nobody enjoys it. So people arrive having thought hard about the job and not at all about the conversation that gets them the job. I'd be glad if more candidates prepared the way an interviewer wishes they would.

What "we" tells me, and the question that breaks it open

The questions I get least out of are the soft-skills ones. Your results. Your achievements. How you work with a team. On paper they're the easy half of the interview. In practice they're where I do most of my digging, because the answers keep arriving in the plural.

People like to say we. We rebuilt the pipeline. We improved how the team communicated. We cut the deploy time in half. I need to know what you did, and "we" is a wall I have to get through before the interview can start.

So I ask the same thing every time. Could you tell me your personal contribution to that process: how did you improve the team's communication?

It's a mechanic rather than a mood. I'm not reading the room when I ask it, I ask it in the same place every time.

Which makes it a thing you can rehearse against. Most people arrive having never once said their own contribution out loud, in the first person.

What the simulation actually reproduces

I built an interview simulator, so treat what follows as a report from someone with an obvious bias, checked against what I do for a living.

The first time I sat through a full session of my own product, two things surprised me. One was how good the model is at asking real questions, not a generic bank but questions built from the actual context: this resume, this vacancy. The other was the voice. It's good enough that you can feel you're in an interview instead of typing at a chatbot.

The follow-ups are the part I'd defend hardest. The model doesn't ask a question, take whatever comes back, and move on to the next one. It's trying to understand the topic. You say the team improved communication, and it wants to know what you did, the same way I would, and then what happened afterwards.

Interview mode exists to close the rest of the gap. You get a random voice, and the model is prompted to behave like an interviewer rather than a helpful assistant: more serious, less accommodating. It works, and you can feel the difference.

And the core constraint is genuinely there. You have to produce an answer out loud, in sentences, with no notes and no editing. That constraint is most of what makes real interviews hard.

Three things it doesn't reproduce

You can tell it isn't a person. That sounds cosmetic and it isn't. Knowing changes how you behave. Nothing is at stake. Nobody is forming an opinion of you that they'll carry into a hiring decision. A good part of what makes a real interview hard is simply knowing that a human being is judging you in real time, and that pressure doesn't survive the knowledge that they aren't real.

You can also cheat, and the cheat is easy. I can think for five minutes and then start answering. In front of a person I have to start roughly now: a couple of seconds to gather myself, then I'm talking, ready or not. That gap is how simulated practice quietly stops working. You take the pause the software allows, you produce a good answer, and you've learned nothing about whether you could have produced it under the real constraint.

And it can't predict your questions. Different companies ask wildly different things, and no simulation anticipates the specific set you'll get on the day.

The laugh you can't have with a machine

The strongest version of the skeptic's case is that you can't reproduce the vibe of a real interview. That's right, and I can tell you exactly what the vibe is.

It's the laugh. There's a moment in a good interview where you both laugh at the same thing and something settles. You understand each other, you can picture the two of you on a team, you're on one line. It's a real signal and it moves real decisions.

You can't be on one line with an AI.

There's a less flattering half to this, and I'll say it plainly since I'm the one on the other side of the table. As an interviewer I'm a human being. I judge people on how they look and how they behave. It's automatic, not a decision I make, which is exactly why no software stands in for it, and exactly why I can't switch it off by deciding to.

The part of the interview that decides whether people want to work with you is not the part you're practising.

Repetition is the whole argument

The reason to practise against a simulation is repetition, not realism.

With an AI you can train against a lot of questions, far more than any person will sit through with you. You can ask it to drill one specific weakness, over and over, at eleven at night, from wherever you happen to be. Distance stops mattering. The skeptic's argument assumes you have an alternative, and for most candidates that alternative doesn't exist: there is no one available to run you through the personal-contribution probe as many times as it takes.

Realism is the wrong thing to judge an interview coaching simulation on. Judge it on repetition, because the problem it solves is access.

How much is enough? It depends on the role, but I'd stand behind two to four weeks per vacancy. Per vacancy, because different jobs pull different questions, and the set you drilled for a platform role isn't the set you'll get for a team lead one.

The objection I hear to this is that repetition makes you sound rehearsed. It doesn't bother me. A polished answer is just a good answer. What it can't be is a lie. It has to come out of real experience, and if I'm not certain it does, I'll keep asking questions until I am.

Which is the honest limit on what practice can do for you. You can rehearse how you tell the story as many times as you like. You cannot rehearse your way into having been more than a participant in the retrospective.

If your interview is tomorrow

Say you have one session left. Stop trying to predict the questions. That's the thing simulation can't do anyway, and it's the worst possible use of your last hour.

Practise the delivery instead. How you think before you answer. How the answer comes out once you've started. What your voice does when you're not sure. That's what a session the night before is for, and it's the layer that transfers completely. It doesn't care which company you're talking to or what they decide to ask.

Don't take the pause the software allows you; start speaking when a real interviewer would expect you to, even when you'd rather have another thirty seconds. And have at least one answer ready about your own contribution to something: first person, with a verb you personally performed. That probe is coming, from me or from someone doing the same job.

You won't get the laugh from a simulation. You'll get everything up to it.