Did they do it themselves? Five questions that show a candidate understood their own work

The reliable way to find out whether a candidate understood the work they sent is to ask them about it. Not about AI, and not about their process in general: about specific choices in their own hand-in. Five questions cover it, and they take about fifteen minutes. The questions are below, with what to listen for.

Why not just detect AI use?

Because it does not work, and it is the wrong question.

Tools that claim to spot AI-written work make mistakes in both directions. Use one as a gate and you will reject honest candidates who write plainly and pass dishonest ones who edit lightly. Signs like "the code is too clean" or "the wording sounds generated" are guesses.

And suppose you could detect it perfectly. A candidate who used an AI tool, checked the result, fixed what was wrong and can explain every part has done exactly what you will ask of them at work. The thing you care about is whether they understood, and that you can test directly.

The five questions

Ask these about the candidate's actual hand-in, with it open in front of you both.

1. "Walk me through what you did first, and why."

You are listening for an order of work that makes sense: they read the material, saw what mattered, then chose an approach. Someone who did the thinking tells you a small story with a wrong turn or two in it. Someone who did not will describe the finished thing instead, because the finished thing is all they know.

2. "Why did you do this part this way and not another way?"

Point at one real decision in their work. Any decision will do: how they split a document, which column they trusted, why the message opens the way it does. A person who made the choice knows what they gave up by making it. A weak answer restates what the work does without ever reaching why.

3. "What would happen if this input were different?"

Change one thing about the material and ask what their work would do. "Suppose the newer policy had no date on it." "Suppose this customer had already been refunded." This cannot be prepared for, and it needs a working picture of the thing they built. Listen for reasoning out loud, not a fast answer.

4. "Where is this wrong, or weakest?"

Everything made in an hour has weak points. People who understand their work know where they are and will tell you without being pushed. Be wary of "I think it handles everything." That is either a person who did not look or a person who does not know.

5. "If you had another day, what would you change?"

This shows whether they can see past what they made. A strong answer is concrete and ranked: this first, because it matters most. It also tells you something no test does: what working with them on a second version would be like.

What does a good answer sound like?

Across all five, the same pattern separates people:

You do not need all five to go well. You need the answers to be about their work, in detail. Detail is hard to fake for fifteen minutes.

How do you make this fair?

Three habits keep it from turning into an interrogation.

• Tell candidates in advance. Say in the brief that you will ask about their hand-in. It is not a trap, and saying so puts off the people who planned to forward an answer.
• Ask everyone the same kind of question. The details differ because the work differs, but every candidate gets the same five.
• Judge the understanding, not the nerves. Give people time. A slow, correct answer beats a quick, smooth one.

Where do the questions come from?

Writing good follow-ups for every hand-in takes time, which is why teams skip this step when there are sixty applicants. Two things make it easier. Give each applicant their own material, so the questions are about something only they worked on. And write the questions from your marking: wherever a candidate lost marks or made an unusual choice, that is your question.

Wooble does this part for you. Each applicant works on material only they have, and for every hand-in you get interview questions built from that person's own work and how it was marked. See how it works, or read what to test now that AI can finish a take-home.