Trust and Hallucinations

Can AI-generated text be reliably detected?

Short answer: Not reliably. Even OpenAI shut down its own AI-text detector in 2023 because of its low accuracy, so treat any detector's score as a hint, not proof.

Last updated Oct 2026 · Sources checked Oct 2026

What OpenAI found

OpenAI's classifier correctly flagged only 26% of AI-written text in its tests and wrongly labeled human writing as AI 9% of the time. It was retired on July 20, 2023 for low accuracy.

Why detection is hard

OpenAI noted detectors are very unreliable on short text, can be fooled by light editing, and sometimes label human writing as AI with high confidence. Very predictable text can't be attributed at all.

What to do instead

Don't accuse someone based on a detector score alone. Ask about the process, look at drafts and version history, or ask the person to explain the work in their own words.

Sources

  1. OpenAI, New AI classifier for indicating AI-written text (updated Jul 2023)

Machine-readable version: /api/answers/can-ai-generated-text-be-detected.json