Testing Pangram AI Detector: Easily Fooled by Sandwiching AI Text

jdjohnson · x · 2026-07-31

A user tested the highly-discussed AI text detector Pangram AI and found its results to be highly inconsistent. The test revealed that if AI-generated paragraphs are sandwiched between human-written openings and closings, the system misidentifies the entire text as "100% human." However, isolating the middle AI paragraphs triggers a "100% AI" detection. The user questioned the meaningfulness of using such tools in practical scenarios.

Related event: AI Text Detectors Easily Fooled by Simple Tricks(3 posts)→

Original post →

More from Apps

Apps channel →