How to spot AI scripts in interviews

Almost every candidate you interview now uses AI to prepare. That is not a problem — we would be worried about the ones who don't. The problem is that a polished, AI-shaped answer looks identical whether the candidate understands it or pasted it thirty seconds earlier.

Screening for "did they use AI?" is the wrong question. The right question is whether the thinking underneath the answer is theirs.

Why banning it fails

Three reasons detection-first interviewing breaks down:

• You cannot enforce it. Remote interviews, second screens, and phone assistants make policing impossible.
• You punish the wrong people. Strong AI-first candidates use assistants constantly and will happily tell you so. Nervous candidates who over-prepare get flagged instead.
• It tests the wrong skill. The job involves AI. An interview that pretends otherwise measures nothing about the work.

So stop detecting and start probing.

Signals that an answer is borrowed, not built

None of these is proof on its own. Two or three together is a pattern.

The answer is structurally perfect but factually thin. Three neat bullets, balanced trade-offs, no numbers, no names, no dates. Real experience is lumpy — it has a specific vendor, a specific week everything broke, a specific number that was wrong.

The abstraction level never drops. Ask about a decision and you get a framework. Ask again and you get the same framework rephrased. People who lived the decision go down a level when pushed, not sideways.

Vocabulary shifts mid-conversation. Casual, ordinary language in small talk; suddenly "holistic stakeholder alignment" when the topic turns technical. The register jump is the signal.

They cannot name the trade-off they rejected. Every real decision has a discarded option. Scripted answers describe the chosen path only.

Timing is inverted. Hard questions get fast, fluent answers; simple factual questions ("who else was on the team?") produce pauses. Recall should be faster than reasoning, not slower.

Follow-ups that separate the two

Have three of these ready for every substantive answer:

1. "What would have to be true for the opposite choice to be right?" — Requires holding the problem in your head, not retrieving a summary.
2. "Who disagreed with you, and what was their strongest argument?" — Scripts have no antagonists.
3. "What did that cost — in money, weeks, or people?" — Real projects have a bill attached.
4. "What would you do differently with the same constraints today?" — Tests whether the experience updated them.
5. "Show me how you'd prompt for this now." — The most useful one. It flips AI use from something to hide into something to demonstrate.

That last question matters more than the rest combined. An AI-first candidate should be able to narrate their assistant workflow: what they delegate, what they verify, where they refuse to trust the output. Someone who pasted an answer cannot describe the process behind it.

The live-work alternative

The fastest way to make scripting irrelevant is to stop relying on recall and start watching work happen.

Give a short, ambiguous problem — 20 to 30 minutes, tools and AI explicitly allowed, screen shared. You are not grading the artifact. You are grading:

• Framing. Do they clarify the ambiguity before producing anything?
• Delegation. What goes to the assistant and what stays with them?
• Verification. Do they check the output, or ship the first draft?
• Recovery. When the model is confidently wrong, do they notice?

A candidate who cannot do the work in front of you cannot do it behind you either. This is the same principle behind our take-home challenge design: a perfect result should be worthless without the process artifact that shows how it was reached.

A short interviewer checklist

• Say out loud that AI use is allowed and expected.
• For each claim, ask for one specific number, name, or date.
• Ask for the rejected alternative.
• Ask them to demonstrate their AI workflow live.
• Replace at least one recall-based question with 20 minutes of observed work.
• Score the process, not the polish.

Where this fits

Interview scripting only feels threatening if your process is built to reward fluent answers. It isn't a threat when you are hiring for leverage — the ability to build systems others reuse — because leverage is visible in behaviour, not phrasing.

That is what our AI-first hiring framework is designed to surface, and what the six pillars of AI-first talent give you language for.