Glossary
AI detector
An AI detector is a tool that claims to tell whether a text or image was made by AI. The honest truth: they are frequently wrong in both directions, and no detector can prove authorship. Treat their scores as a hint, never as evidence.
Paste an essay into a detector and it returns a verdict like “87% likely AI-generated.” That number looks scientific. It isn’t. Detectors guess from statistical patterns, and AI text is specifically optimized to look like human text, so the patterns keep dissolving.
The errors run both ways. Lightly edited AI text routinely passes as human. Meanwhile, real people get flagged, non-native speakers and writers with very regular styles most of all. OpenAI shut down its own text detector back in 2023 over low accuracy, which says plenty. Students have faced accusations over false positives; that’s the real harm of treating these scores as proof.
So what works instead? For images, provenance labels (like C2PA content credentials) beat after-the-fact guessing. For text, conversation does: someone who wrote a piece can discuss it. If you use a detector at all, use it as one weak signal among several, never as the verdict.
Where you’ll meet this
GPTZero, Turnitin’s AI flag, Copyleaks and similar tools in schools and editorial workflows; “AI-generated” labels on social platforms (increasingly via embedded watermarks like image credentials, which are more reliable than text detectors).