Interview questions that expose AI-assisted answers, and stay fair to everyone

The interview questions an AI tool cannot answer for a candidate are the ones built from the candidate's own history and from what happens live in the room: "Which line of your resume would you most like to go deep on? Who signed it off, and what broke first?" or "Here is a situation our team had last quarter. What do you check first? Now the budget is halved. What changes?" An assistant can write a fluent general answer; it cannot know who signed off, or react well to a constraint it has not heard yet. Below are 24 questions grouped by technique, each with what a real answer contains and what an AI-shaped answer tends to sound like, plus a scoring rubric.

How to tell, during the conversation, that an answer may be generated is covered in how to detect AI-generated interview answers, and how the tools themselves work is in AI interview cheating tools explained. This page is the question set.

Why do these questions work, and why are they fair?

Each question below does at least one of three things a tool handles badly: it asks for details only the person who did the work would know, it introduces information the candidate has just heard, or it requires back-and-forth. None of them is a trick. They are ordinary structured-interview practice, which is why they are fair: a prepared candidate with real experience does well, whoever helped them prepare. Structured interviews also predict job performance better than most other methods; Sackett and colleagues' 2022 re-analysis in the Journal of Applied Psychology still ranked them first among the selection procedures it reviewed.

Questions anchored in their resume

Write these the night before from the candidate's own resume. A generic tool has never seen these details.

Questions where the candidate chooses the example

A prepared script cannot anticipate which example the candidate will pick, and a candidate reading generated answers often picks the example that fits the script rather than their best one.

5. "Pick a piece of work from the last two years you'd be happy to go deep on for ten minutes. Start wherever you like."
6. "Tell me about something you built or changed that is still in use. What would the people using it complain about?"
7. "Pick a decision you made that turned out wrong. How did you find out?"
8. "What's something you know a lot about that isn't on your resume?"

Then stay with the chosen example. The follow-ups below, asked about the same story, are where the score comes from.

Follow-ups that need the candidate's own memory

Ask at least two of these on every main example. Real experience gets more detailed with each one, including details nobody would invent; generated or borrowed material gets vaguer, repeats itself or retreats to principles.

Scenarios you supply, with a twist

Describe a short, real situation from your own team, then change one constraint halfway through. Context the candidate has just heard is hard to prepare for, and the change forces thinking rather than retrieval.

15. "Last quarter, our [team] saw [problem, two sentences]. You've just joined. What do you look at first?" Then: "The data you asked for doesn't exist. What now?"
16. "A client wants [deliverable] by Friday; the team says it takes three weeks. What do you say to each of them?" Then: "The client's boss calls you directly. What changes?"
17. "Here's a one-page summary of a decision we made. What would you have asked before agreeing to it?"
18. "You have two good candidates and one offer. Here's what we know about each. Which questions decide it?" (for recruiting or management roles)

What to listen for: questions back to you, a sensible first step, and an honest "I'd need to know X." A generated-sounding answer covers every angle evenly, asks you nothing, and treats the twist as a new essay prompt.

Live work and teach-back

19. "Here's a paragraph from one of our customer emails. Talk me through how you'd rewrite it, as you edit." (Share a document; watch the editing, not the final text.)
20. "Here's a small bug in our code. Think out loud while you look." (For technical roles; the talking matters as much as the fix.)
21. "Explain [concept from their field] the way you'd explain it to a new hire on their first day."
22. "Now explain it to the CFO in two sentences."

Callbacks across the interview

23. "Earlier you mentioned [detail from answer 5]. How did that affect [later project]?"
24. "You said [person] disagreed with you. Did that change how you handled the next project with them?"

A real career is connected; details from one answer show up in another. Pre-written answers rarely refer to each other. Note one detail from each main answer so you can call back to it later.

How should you score the answers?

Score the follow-ups, not the opener. A polished first answer with nothing behind it should not score well, and a halting first answer that fills with detail under questioning should. Use anchors like these, adapted to the competency:

Write down what the candidate said, not what you suspect. "Could not name the system or anyone involved after two follow-ups" is evidence; "probably used AI" is a guess. More anchors are in behavioral interview questions with a scoring rubric.

Which questions should you drop?

• Trivia and definitions ("What is a primary key?"). Any tool answers instantly, and the answer says nothing about the job.
• Stock behavioral prompts asked without follow-ups ("Tell me about a time you showed leadership," then the next question). They invite the exact answer a tool writes best.
• "Greatest weakness" and other rehearsed classics. Coached answers were common long before AI.
• Brainteasers. Unrelated to the job, and they disadvantage nervous candidates.
• "Are you using AI right now?" Nobody can answer it usefully, and an honest candidate will remember being asked.

A 30-minute plan that uses these questions

0-3 min    Welcome; state the rules ("your own words, notes you
           prepared are fine, no live help")
3-13 min   Candidate-chosen example (Q5) + three follow-ups (Q9-Q14)
13-20 min  One resume-anchored question (Q1-Q4) + two follow-ups
20-27 min  Your scenario with a twist (Q15-Q18)
27-29 min  One callback (Q23-Q24)
29-30 min  Their questions; next steps

This needs only four main questions; the follow-ups do the work. If you use Interview Signal, which we make, the question guide it builds from the job description keeps must-ask questions pinned until they are covered, and its suggested follow-ups carry a reason (probe, clarify or evidence), so you can keep your attention on the candidate rather than the list.

How do you keep these questions fair for every candidate?

• Ask the same core questions of every candidate for the role; vary only the resume-anchored details.
• Allow thinking time. A pause before a hard question is normal and is not evidence of anything.
• Agree accommodations in advance. Some candidates use notes, captions or extra processing time; see ADA accommodations in interviews.
• Judge content, not polish. Second-language speakers may answer more formally or slowly; score the detail, not the fluency. How to interview bilingual candidates has more.
• Tell candidates the rules first. The wording for invitations is in candidates using AI in interviews: a fair policy.