Author Finds AI Detectors Rate Their Pre‑ChatGPT Writing as Only 5% Human

The author evaluated eleven AI detection services on writing produced before ChatGPT existed. The tests revealed the writing was scored as only five percent human. The result suggests the detectors often flag older text as AI‑generated. The experiment was shared via a Substack post. The author’s findings raise questions about detector reliability. The post highlights challenges in distinguishing human from AI content. The analysis does not name specific detectors. The discussion invites further scrutiny of detection methods.