Skip to main content

How to Use AI Evaluations to Detect Interview Fraud

This guide covers how to use AI to analyze candidate interviews for the telltale signs of AI usage.

Written by Melisa Cooper

Overview

With the explosion of hidden screen overlays, real time audio lip syncing, and generative AI interview whisperers, AI interview fraud is at an all time high. Recent industry data reveals that nearly 40% of candidates in technical and operational interviews admit to using AI assistance during interviews.

Because generative AI knows exactly how to ace interviews, candidates who use it often look flawless on paper and rank as top tier talent. To protect your pipeline, you need to combine automated flags with a human check.

AI Evaluation prompts

Copy and paste these exact prompts (parts in bold) into your AI Interviews > Evaluation Categories to flag AI generated candidate responses. See How to set up AI Interviews.

  1. Clinical scripting (rank Weak)

    1. What to look for: Does the answer read like a sterile corporate checklist, rather than a real person speaking naturally about their career?

  2. Skill padding (rank Weak)

    1. What to look for: Does the candidate claim elite mastery over a massive, unrelated pile of hard and soft skills without showing a realistic learning curve?

  3. Buzzword stuffing (rank Weak)

    1. What to look for: Does the candidate pack their answer with technical buzzwords and tools without actually explaining the process of how they used them?

  4. Overly formal transitions (rank Weak)

    1. What to look for: Answers that lack natural conversational pauses (like "um" or "uh") and instead rely on overly formal transition words like "furthermore," "moreover," or "in conclusion."

Analyzing the results

Scores are given on a stoplight basis:

🟢 Green indicates low fraud detection

🟡 Yellow indicates mild fraud detection

🔴 Red indicates high fraud detection

Human verification checks

Human reviewers should audit the interview audio using these specific real world telltales:

  • Audio Lag and Latency: Listen for strange, mechanical micro pauses or a slight delay during rapid back and forth questions. This usually means a voice synthesizer or live translator is working in the background.

  • Robotic Pacing: Watch out for a perfectly flat, uniform cadence, or long, awkward silences right before they suddenly unleash a flawless, rapid fire technical answer.

  • Keyboard Cues: Listen closely during pauses. If you hear faint, furious typing right after you ask a question, they are likely feeding prompts to a live AI.

  • Interrupting the Script: Pay attention to how they handle mid sentence interruptions. Real people will organically pivot when interrupted. Someone relying on an AI whisperer will freeze up or completely lose their place while waiting for the AI to give them a new response.

Keep in Mind (The Technical Limits)

Before rolling these out, make sure your team understands what a text based AI evaluator cannot do:

  • It cannot read the job description: The evaluation tool only reads the interview transcript. It does not know what was in your original job posting, so it cannot tell if a candidate is just copy pasting your own requirements back to you.

  • It is a smoke detector, not a judge: These flags are meant to point out weird text patterns and save your team time by highlighting who to review. They should not be used to automatically disqualify anyone without a human checking the audio first.

Did this answer your question?