How accurate is gptzero?

GPTZero, one of the more widely known AI detection tools, has shown mixed accuracy results across various independent tests and studies, generally performing reasonably well on clearly AI-generated text that hasn't been edited, but struggling more with edge cases. Its detection method relies primarily on analyzing two key metrics, perplexity, which measures how predictable or surprising the word choices in a text are, and burstiness, which looks at variation in sentence length and structure, since human writing tends to be more varied and less uniform than typical AI output. Studies have found that GPTZero's accuracy can drop significantly when text has been lightly edited or paraphrased by a human after being AI-generated, which is an increasingly common workflow that makes detection genuinely difficult across the entire industry, not just for this specific tool. False positive rates are also a real concern, with some non-native English speakers and very structured, formal writing styles occasionally getting flagged incorrectly as AI-generated. GPTZero's developers have continued updating and refining the tool over time, and it generally performs comparably to other leading detectors like Originality.ai and Copyleaks, but no current AI detector, including GPTZero, should be treated as definitively reliable for high-stakes decisions.