GPTZero flagged your essay, but you wrote it yourself — here's what that means

GPTZero is one of the most widely used AI detectors in schools. If it has marked an essay you wrote yourself as AI-generated, you are not the first — and the reason is usually the way the detector measures text, not anything you did wrong. This guide explains how GPTZero reaches its verdict, why non-native English is especially exposed, and what to do next.

How GPTZero scores text

GPTZero has described two signals at the heart of its approach (GPTZero):

• Perplexity — how surprising each word is to a language model. Low perplexity means the model could easily have predicted your next word.
• Burstiness — how much that predictability varies across sentences. Human writing tends to mix short and long, plain and unusual sentences; AI text is often evenly smooth.

GPTZero says its current model uses these as some of several indicators, alongside a deep learning classifier. The output is a probability that the text was AI-written — not a record of how it was written.

Why ESL writing looks “predictable”

When you write in a second language, you tend to choose words you are sure of and sentence patterns you have practised. That is good writing strategy — but it produces exactly the low-perplexity, low-burstiness profile detectors associate with AI. In the Stanford study of seven detectors, human-written TOEFL essays were misclassified as AI 61.22% of the time on average, while essays by US eighth-graders were classified almost perfectly (Liang et al., 2023).

The researchers also found that when the same essays were rewritten with richer, more varied word choices, far fewer were misclassified. The problem is the style of simple English, not the honesty of the writer. We explain the mechanism in depth in why AI detectors flag human writing.

How accurate is GPTZero?

No detector is accurate enough to serve as proof. An independent evaluation of 14 tools, GPTZero among them, found every one scored below 80% accuracy (Weber-Wulff et al., 2023). OpenAI withdrew its own classifier after it mislabeled human text as AI 9% of the time (OpenAI, 2023). See every figure in our AI detector statistics roundup.

What to do if you're flagged

1. Stay calm and ask for the result. Ask which tool was used, what score it gave, and which passages were highlighted.
2. Collect your process evidence. Google Docs or Word version history, earlier drafts, outlines, notes, browser history for your sources.
3. Offer to explain your work. Walking your instructor through your argument, or writing a short passage under supervision, tests authorship directly.
4. Cite the research. Point to the documented bias against non-native writers — calmly, as context, not as an attack.
5. Put it in writing if needed. Our appeal letter template gives you a structure.

The full process is in falsely accused of using AI? What to do.

Lowering the risk next time

Write in a tool that saves version history, keep your notes, and revise the sentences that sound translated. Varying your sentence rhythm and adding concrete details from your own experience is better English and less “predictable” to a model. Our seven habits that prevent false flags go further.