A frustrated author recently posted a simple plea on Hacker News: “No, I swear I wrote this.” The comment, attached to a discussion about AI-generated text, captured a growing frustration among writers who find themselves falsely accused of using artificial intelligence tools. As detection software becomes more common, so do false positives that mistake human prose for machine output.
The Rise of False Positives
AI detection tools such as GPTZero and Originality.ai have gained popularity among educators, publishers and editors seeking to identify machine-written content. These tools analyze patterns like perplexity and burstiness to flag text that resembles output from large language models such as OpenAI’s GPT-4 or Anthropic’s Claude. However, the same statistical signals can appear in human writing, especially in technical or academic contexts where clarity and structure are valued.
Several high-profile cases have emerged in recent months. Authors have reported being rejected from literary magazines after their submissions were flagged as AI-generated. Students have faced academic discipline for essays they wrote themselves. In each case, the accused must prove a negative, often with little recourse.
Why Current Detectors Fall Short
Detection algorithms rely on assumptions about how humans and machines write differently. Those assumptions break down in practice for several reasons:
Researchers at Stanford University found that popular detectors misclassify nearly half of human-written texts as AI-generated when the writing is edited for clarity. The margin of error makes these tools unsuitable for high-stakes decisions.
Why This Matters
The consequences extend beyond individual embarrassment. Publishers risk rejecting authentic voices. Educators may penalize honest students. And the broader trust in written communication erodes when every sentence is treated with suspicion. Writers now face an invisible burden: proving their own humanity in an age where machines can mimic it convincingly.
Until detection technology improves or alternative verification methods emerge, the burden will remain on the accused. For now, many are left repeating the same refrain: “No, I swear I wrote this.”



