Authors Guild test finds some AI detectors perfectly identify human writing while others fail on every single text
AI detectors are a coin flip. Pangram got it right 100% of the time. ZeroGPT got it wrong 100% of the time.
Why it matters
AI detector reliability is broken and inconsistent, creating liability for publishers, platforms, and writers relying on these tools to distinguish human from machine content. The core problem: language models trained on human writing make that distinction statistically meaningless.
The key facts
13 to knowAuthors Guild conducted detector accuracy test on human-written texts
Pangram: 100% accuracy identifying human writing
Grammarly: 100% accuracy identifying human writing
Sidekicker: failed on all tested texts (flagged human writing as AI)
ZeroGPT: failed on all tested texts (flagged human writing as AI)
Root cause: LLMs trained on professional human writing, creating statistical similarity between human and AI output
Published June 25, 2026
Authors Guild tested 5 AI detectors on human-written texts
Pangram: 100% accuracy identifying human text
Grammarly: 100% accuracy identifying human text
Sidekicker: flagged human text as AI-generated
ZeroGPT: failed to identify human text (flagged as AI)
Root cause: language models trained on professional human writing create statistical similarity between human and AI output
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: The Authors Guild tested five AI detectors on human-written texts. Pangram and Grammarly correctly identified all of them, while Sidekicker and ZeroGPT flagged human-written articles as AI-generated. But the Guild also warns of a paradox: professionally written texts look statistically similar to…