How to Use GPTZero: A Step-by-Step Walkthrough

How to Use GPTZero, Step by Step

GPTZero was built by Princeton student Edward Tian in early 2023 and has become one of the detectors most commonly used by educators. The walkthrough below assumes you want to run a document check on the public web app. Classroom and LMS integrations work the same way under the hood.

1. Open GPTZero

Go to gptzero.me. You can run a short check without an account. For longer documents or to keep a history, sign up with an email or a Google account.

2. Paste text or upload a file

The main input accepts raw text. GPTZero also supports uploading .pdf, .docx and .txt files on the free tier. Paste a sample of at least a few paragraphs for a useful result; shorter samples produce less confident scores.

3. Click Get Results

The analysis runs in a few seconds for short samples and takes longer for full documents. GPTZero returns three things:

• A probability-style score. A number between 0 and 100 that reflects the model's confidence that the passage is AI generated.
• A short verdict. The score is grouped into bands: human, mixed and AI.
• Sentence-level highlights. Each sentence is colour coded to show how AI-like the model considers it.

4. Read the sentence highlights, not just the top number

The single top-line score is a summary. The sentence highlights are where the useful information lives. If 90 percent of the sentences look human and one paragraph lights up red, the useful question is what is different about that paragraph, not whether the whole document is "AI".

5. Export or share

GPTZero can produce a PDF of the result for records. Instructors share these in integrity conversations. If you are the writer and want to understand a flag, read the report yourself before responding.

What the GPTZero Verdict Actually Means

GPTZero's current documentation describes its detection as a deep-learning classifier trained on paired human and AI examples. The company's early marketing emphasised two numbers, perplexity and burstiness. GPTZero now says those metrics are no longer the primary signal and that the current model learns its own features during training. See GPTZero's own methodology page for the current description.

The verdict bands are useful shorthand but should be treated as a probability, not a certainty:

• A human verdict means the model saw mostly patterns associated with human writing in its training set
• A mixed verdict usually means part of the document reads as AI-like and part reads as human-like; this is common on heavily edited AI drafts
• An AI verdict means the model is highly confident the passage was generated by a large language model

Independent studies, including Liang et al., 2023, have shown that detectors can produce false positives above 60 percent on text written by non-native English writers. GPTZero was one of the detectors evaluated. Short samples and heavily edited text also lower reliability.

Good Habits When You Use GPTZero

• Check longer samples, not single sentences, when possible
• Compare the sentence highlights to the actual writing to understand what the model is flagging
• Avoid treating a single detector score as proof of anything on its own
• If you are a student or writer and have been flagged, document how the piece was written and ask for a conversation rather than a verdict

If You Need to Rewrite an AI Draft

If your goal is to clean up an AI draft so it reads like you wrote it, use a humanizer and then edit the output by hand. Walter Writes offers a 300-word trial with no card required at walterwrites.ai/ai-humanizer. Voice and audience controls tend to produce more natural results than word-level synonym swaps. Verify the current trial terms on their site before relying on it.

Sources

• GPTZero. How GPTZero Works: current methodology description.
• Edward Tian. GPTZero: an AI text detector: tool home page.
• Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., Zou, J. GPT detectors are biased against non-native English writers, 2023.