Can AI Content Be Detected? What the Evidence Says
June 25, 2026 · 7 min read
Detectors are unreliable enough that acting on them alone is unfair to real writers.
High false-positive risk
Non-native English writers are flagged disproportionately by most detectors.
Easy to defeat
Light editing or paraphrasing collapses detector confidence.
Better approaches
Process evidence — drafts, version history, oral defense — is far more robust.
Policy over policing
Clear rules about permitted use beat an arms race nobody wins.