To bypass GPTZero, the text has to read like a person wrote it: plain words, sentences of different lengths, specific details, and none of the habits AI models share. Two things get you there. Write it with a model trained to write like a person, or rewrite the AI text you already have with a humanizer.
Hemmingway does both, with Hemmingway-1.5, the undetectable AI. In our test, GPTZero read 90% of what Hemmingway-1.5 wrote as human, at every thinking level, with each text run once.
What GPTZero measures
GPTZero is an AI detector. You paste in text and it gives back how likely it is that an AI wrote it. Its site says it flags text from models including ChatGPT, Gemini and Claude.
It started out with two measures, which it calls perplexity and burstiness. Perplexity is how predictable each word is to a language model. Burstiness is how much that predictability changes across the text. AI text tends to be predictable all the way through, and a person's jumps around. GPTZero has since moved to a deep-learning model, and its own help pages say the two measures are still part of what it looks at.
Either way, it reads patterns in the text. It cannot see who typed it.
Why ChatGPT-style text gets flagged
The big chatbots write with the same habits, and those habits are the patterns GPTZero picks up:
- Even rhythm. Every sentence about the same length, every paragraph about the same size.
- Stock words. "Moreover", "delve", "crucial", "it's worth noting", "I hope this finds you well".
- The wrapper. An opening that restates the ask, and a closing that sums up or offers more help.
- Nothing specific. General lines where a person would give the name, the day or the number.
In our test, the answers that six of the biggest models gave to 80 everyday asks passed GPTZero and four other detectors between 13% and 26% of the time. GPT-6 Astra, the ChatGPT model we tested, passed 20%.
What works
Write it with Hemmingway-1.5
- Open Hemmingway and ask for what you need in plain words: the essay, the email, the post.
- Give it your facts and points. It builds the text on them.
- Paste a paragraph you wrote if you want it in your voice.
- Use the text. It comes back on its own, with no preamble and no "let me know if you'd like changes".
GPTZero read 90% of Hemmingway-1.5's texts as human at each of the four thinking levels: off, low, medium and extra high. The free tools work the same way, for example the AI essay writer and the AI email writer.
Humanize the text you have
Paste it into the free AI humanizer, say who will read it and how it should sound, and it rewrites it in a person's voice, keeping what it says. By its Hugging Face card, 89% of Hemmingway-1.5's rewrites of AI text passed GPTZero and four other detectors. How to humanize AI text goes through it step by step.
Fix it by hand
Cut the opening and closing lines, swap each general line for the specific fact behind it, vary the sentence length and drop the stock words. How to make AI text sound human has the full list.
What does not work
Swapping words for synonyms. The rhythm, the wrapper and the missing details stay, and those are what GPTZero reads.
Our GPTZero numbers
| Thinking level | Hemmingway-1.5 passes GPTZero |
|---|---|
| Off | 90% |
| Low | 90% |
| Medium | 90% |
| Extra high | 90% |
That is 1,848 texts in 13 kinds of writing, each run through GPTZero once, with nothing picked out and nothing run again. Against GPTZero and four other detectors together, 90% passed, more often than the 86% that real human writing got. Does AI writing pass GPTZero? has the results next to other models, and our AI detector test has the whole method.
Try HemmingwayDownload the app
Common questions
Can you bypass GPTZero?
Yes. GPTZero read 90% of Hemmingway-1.5's writing as human in our test, at every thinking level. Writing it with a model trained to write like a person, or rewriting AI text so its habits are gone, is what does it.
Does GPTZero detect ChatGPT?
Its site says it flags text from ChatGPT, Gemini and Claude. In our test, GPT-6 Astra, the ChatGPT model we tested, passed GPTZero and four other detectors together only 20% of the time.
Does a humanizer get past GPTZero?
Hemmingway's does, nearly nine times in ten. By its Hugging Face card, 89% of Hemmingway-1.5's rewrites of AI text passed GPTZero and four other top AI detectors. A rewriter that only swaps synonyms leaves the patterns GPTZero reads.
Does GPTZero flag human writing?
Yes. In our test, GPTZero and four other detectors together flagged 14% of the 1,670 texts people wrote themselves. A detector reads patterns, not who typed the words.
Does Hemmingway pass Turnitin?
Turnitin's detector is not open to the public, so it was not in our test. Our test used GPTZero and four open detectors, Fakespot, Vanguard, WasItAIGenerated and MELD, and 90% of Hemmingway-1.5's writing passed them on the first try.