What interview questions catch real-time AI use without any software?
By Pinal Dave | Last updated: 2026-08-01
TL;DR: Ask for improvisation an AI cannot pre-script: a follow-up about a personal decision behind the code, a request to change the approach mid-answer, or a question about a recent, obscure detail of their own resume. These techniques help but have limits, which is why they work best alongside, not instead of, technical proctoring.
The short answer
Before any software existed to detect AI-assisted interviewing, experienced interviewers relied on conversational technique to surface the same problem: is this person answering from real understanding, or repeating something generated elsewhere. Those techniques still work and cost nothing to add to an existing interview process, but they depend entirely on interviewer skill and attention, which does not scale reliably across a large hiring team or high interview volume.
The core idea behind all of these techniques is forcing improvisation. A large language model can produce a fluent answer to a known question instantly. It is far less useful for a candidate who has to defend a specific personal choice, adapt a solution live, or explain something only they would know about their own background.
The evidence
- Karat: 80% of candidates use LLMs during coding tests that ban them, meaning interviewers should assume some baseline rate of AI use even when it is prohibited.
- Fabric (19,368 interviews): 38.5% of candidates flagged for AI-cheating behavior across a large sample, underscoring that this is common enough to prepare for deliberately, not treat as a rare surprise.
- Greenhouse survey (4,136 respondents): 91% of hiring professionals encountered suspected AI-generated answers, suggesting most interviewers already sense this happening even without a formal technique for surfacing it.
Question techniques and what they catch
| Technique | What it forces | What it does not catch |
|---|---|---|
| "Why did you choose this specific approach over the alternatives?" | Personal reasoning behind a decision | A well-coached candidate rehearsing likely follow-ups |
| Ask to change the solution live, mid-answer | Real-time adaptability | Whether an AI tool is generating the adapted answer too |
| Ask about a specific, obscure resume detail | Genuine personal history | Nothing reliably, if the resume itself was fabricated |
| Request the answer in a different format, like a diagram or verbal walkthrough | Understanding beyond memorized text | A second person coaching the response off-screen |
Step-by-step: building this into interviews
- Prepare one or two improvisation follow-ups per core question, not as gotchas, but as a natural part of a technical or behavioral round.
- Watch response timing, not just content. A long pause before a suspiciously fluent, complete answer is a soft signal worth noting.
- Ask the same candidate to explain their own answer differently. Genuine understanding usually survives rephrasing; a memorized or fed answer often does not.
- Do not rely on this alone for high-stakes decisions. These techniques are useful signals, not proof, and should be combined with identity and behavior proctoring for anything consequential.
- Debrief interviewers on patterns they notice. Sharing what worked across a hiring team improves detection more than any single interviewer working in isolation.
FAQ
Can a skilled candidate defeat these techniques by preparing for them? Some can, especially if they know the interviewer's style in advance. This is exactly why these techniques work best as one layer, not the only layer, of a detection strategy.
Do these questions risk making the interview feel adversarial? Framed as normal follow-up curiosity rather than an interrogation, most candidates do not perceive them as adversarial, since genuine experts enjoy discussing their own reasoning.
Is this a substitute for AI proctoring software? No. These techniques catch some cases through skilled human judgment, but they do not verify identity, detect a second voice, or provide timestamped evidence if a decision is later questioned.
Do these techniques work equally well for coding, sales, and other non-technical roles? Yes, the underlying principle, forcing improvisation and personal specificity, applies across role types, though the specific questions differ.
How much interviewer training does this actually require? Very little to start. A short internal guide with example follow-up questions per role type is usually enough to raise detection meaningfully across a hiring team.