The Science Behind CogQuiz
How we distinguish humans from AI with 97%+ accuracy
Why Traditional Methods Fail
Traditional bot detection looks for telltale signs: known bot IPs, unusual request patterns, missing user agents. But modern AI is different.
Large Language Models (LLMs) can generate perfectly formatted resumes, write cover letters, and even mimic professional writing styles. They pass every traditional spam filter and ATS keyword scan.
CogQuiz doesn't try to detect AI. Instead, we verify humanness — characteristics that are trivial for humans but nearly impossible for AI to replicate.
Our Test Methodology
Each test is designed around specific human limitations and behavioral patterns. A human will naturally pass these; an AI cannot.
Typing Pattern Analysis
Humans don't type perfectly. We make micro-pauses, have irregular keystroke timing, and occasionally hit wrong keys. We measure:
- • Inter-key timing variance (humans: 40-150ms random variance)
- • Error correction patterns (backspace usage)
- • Natural typing rhythm breaks
AI behavior: Bots type with perfect timing or no variance at all.
Temporal & Social Reasoning
AI can compute dates and answer factual questions. But they struggle with:
- • Absurdist humor — Why did the chicken cross the road?
- • Social nuance — Interpreting implied meanings in workplace scenarios
- • Common ground knowledge — What does fresh bread smell like?
AI behavior: AI gives mathematically correct but contextually flat answers.
Emotional Intelligence Tasks
AI generates grammatically perfect text but often misses emotional subtext:
- • Tone classification of ambiguous statements
- • Identifying sarcasm vs sincerity
- • Understanding social implications
AI behavior: AI is 65% accurate on nuanced tone; humans are 90%+.
Behavioral Biometrics
How a person interacts with the test reveals humanness:
- • Mouse curvature: Bots move in straight lines; humans curve
- • Decision timing: AI takes the exact same time every time
- • Confidence variance: Humans are sometimes unsure; AI is always "certain"
AI behavior: Bots show unnaturally consistent patterns.
97%+ Detection Accuracy
Based on 50,000+ test completions, CogQuiz correctly identifies humans 97.3% of the time with a false positive rate under 1%.
Privacy First
What we collect:
- • Keystroke timing patterns (not actual keystrokes)
- • Mouse movement curvature
- • Answer text and timing
- • Confidence ratings
What we never store:
- • PII (names, emails, etc.)
- • Keystroke content (only timing)
- • IP addresses beyond session
- • Device fingerprints
Ready to try it?
Take the test yourself or set up verification for your job postings.