The One Question That Exposes AI-Coached Answers in Under 30 Seconds

AI-coached answers sound fluent, polished, and structured.
But they collapse the moment you ask this one question:

“What changed?”

This question instantly reveals whether the candidate:

• Has real, lived experience

• Or is reciting an answer generated by ChatGPT or a coaching service

Because real experience contains adaptation, not just information.

Why AI-Coached Answers Sound Convincing

Modern AI can generate:

• Perfectly structured frameworks

• Confident tone

• Clean bullet points

• Process sequences that sound senior

But it cannot generate lived adaptation because:

• Real-world work contains inconsistencies

• Decisions evolve with constraints

• Priorities shift mid-project

• Tradeoffs are imperfect and contextual

As highlighted in Artificial Intelligence and the limits of reason (2025), AI reasoning is ultimately statistical pattern-matching. It cannot step outside its training data to handle real-world unpredictability or novel edge cases.

AI answers are straight lines.
Real experience is messy.

The Question That Breaks the Script

After the candidate gives a polished answer, say:

Thank you. What changed?

Then pause.

Let silence do the work.

Why This Works

Real contributors can describe:

• Shifts in requirements

• Mistakes they corrected

• Blockers they handled

• Surprises that forced adaptation

• Tradeoffs that were not obvious upfront

AI-sourced answers cannot.

Because the AI is describing:

• A textbook solution

Not an actual event

How to Use the Question in Different Interview Types

For Behavioral Interviews

Candidate says:

I led the migration to microservices.

Ask:

Thank you. What changed during the migration that required you to adjust your approach?

If they cannot describe:

• New constraints

• Resource shifts

• Performance surprises
  They likely did not lead it.

For System Design

The candidate explains a scaling architecture.

Ask:

Great. At what point did your initial design assumptions stop being true?

These tests:

• Real-world constraint handling

• Production incident awareness

• Contextual reasoning

No AI tutorial includes this.

For Coding Interviews

The candidate produced the correct code, but it seems rehearsed.

Ask:

If the input size increased by 100 times, what is the first thing you would need to rethink?

A real developer will:

• Think aloud

• Describe complexity tradeoffs

• Adjust algorithmic choices

AI-coached responses stall or loop.

The Signal to Look For

The candidate should be able to:

AI answers show:

• No timeline

• No conflict

• No pivot

• No personal involvement markers

The Follow-Up Ladder (If Needed)

If they stall, go here:

What was the hardest tradeoff?

If still scripted:

Who pushed back on your approach?

If still generic:

What did you change after seeing the outcome?

Three questions.
AI fails on all three.

The Documentation Template (Audit Safe)

Candidate provided structured responses but was unable to describe how the approach evolved or changed in response to constraints or new information. This suggests limited direct ownership of the work discussed

No accusation.
Just signal.

Conclusion

You do not need:

• Gotcha questions

• Trick puzzles

• Aggressive interrogation

You only need:

What changed?

Because real experience changes.
Scripts do not.
AI does not.
Proxies do not.

This technique:

• Detects fraud fairly

• Reduces bias risk

• Improves hiring accuracy

• Works in any industry or role level

It is a simple question with strategic impact.