Technical interviews in 2025 increasingly include:

• AI-assisted coding

• Copied system design templates

• Coached project narratives

• Proxy interviewers interacting on behalf of candidates

Traditional interview methods were not designed to detect this. But new detection tools can surface authenticity signals in real time.

Tools to Detect Plagiarism in Technical Interviews

We review the leading approach and compare it to older manual methods still used today.

1. Sherlock (AI Interview Integrity Layer)

Sherlock is purpose-built to detect fraud, impersonation, and AI-coached responses in live interviews conducted on Zoom, Teams, Meet, or browser-based coding environments.

What Sherlock Detects Automatically

Why This Works

Sherlock does not try to guess intent.
It detects behavioral and reasoning patterns that differ between:

• A person actively thinking

• A person reading or relaying generated answers

In a multi-company dataset across 14 roles, candidates who could not adapt reasoning when constraints changed were 7 times more likely to have used coached or AI-assisted answers.

Sherlock exposes that adaptability gap in real time.

2. Manual Screen Recording Review

Still common in many companies.

Pros

• Useful when detecting obvious fraud

• Can be used during audit review

Cons

• Time consuming

• Reviewer bias risk

• Misses subtle coaching

• Only works after the interview is already over

Manual review is reactive, not preventive.

3. Plagiarism Checkers for Code

Typical tools:

• GitHub Copilot Output Detection

• MOSS (Measure of Software Similarity)

• Codequiry

• JPlag

What They Detect

• Structural similarities in code

• Common template reuse

• Shared logic flow patterns

Limitations

• They require final code, so they cannot detect help during the interview

• They do not evaluate reasoning or decision-making depth

• They miss coached verbal explanations entirely

Code plagiarism detection helps with output authenticity, not thinking authenticity.

4. Live Pair Programming Observation

Interviewers watch:

• How candidates break down problems

• Where they pause

• How they adjust if stuck

This is effective, but only when interviewers are trained to ask:

1. What alternative approaches did you consider

2. What changed while solving

3. What tradeoffs influenced your decision

Untrained interviewers mistake fluency for competence.

5. Behavioral Cross-Checking

Verification technique:

Ask:

What changed in the project after the first release?

Ask again later in the interview in different words.

Authentic candidates:

• Answer consistently

• Provide detail tied to real constraints

Coached candidates:

• Repeat surface-level phrases

• Produce generic improvement narratives

• Cannot anchor the story to actual technical tradeoffs

This reveals experience ownership, not memorized storytelling.

6. Constraint Shift Test (High Signal)

This is the single most effective manual method.

After the candidate explains a system or code approach, ask:

If your assumption about traffic, latency, or data shape changed, what would have to be redesigned?

Real engineers update their solution clearly.
AI-generated or coached answers collapse when conditions change.

This is the core reasoning integrity check.

Summary Comparison

Conclusion

Detecting plagiarism in technical interviews is not about:

• Spotting suspicious facial expressions

• Asking harder questions

• Lengthening interview loops

It is about evaluating how the candidate thinks when the situation changes.

Tools like Sherlock surface this automatically.
Traditional methods require interviewer training and more time.

To scale fairly and reliably, companies benefit from combining:

• Real-time detection

• Reasoning-based interviewing

• Clear documentation standards

This reduces:

• Hiring risk

• Performance failures in the first 90 days

• Legal and audit concerns