What GPTZero Measures, and What a Rewrite Changes
GPTZero is a third-party classifier we do not run, are not affiliated with, and cannot speak for. What we can do is explain what it is documented to measure, and be precise about which of those things a rewrite changes.
Your AI draft
Humanized output
Detector check includedRun the humanizer to see detector results.
Detector verification status
Perplexity, in plain terms
Perplexity measures how surprised a language model is by the next word, given everything that came before it in the sentence. A low perplexity score means the words that follow are the ones the model would have guessed anyway — highly predictable, in the narrow sense of matching what a language model expects language to do. Because a model like ChatGPT is trained and optimised to produce exactly that kind of likely, well-formed continuation, its output naturally sits in a narrow predictability band across a whole passage. Sustained human writing rarely stays in that band for long, not because people try to be unpredictable, but because real thinking produces the occasional odd word choice, the aside, the sentence that does not follow the expected pattern. Low, steady perplexity across several paragraphs is the specific thing GPTZero is documented to measure here — not tone, not grammar, not correctness.
Burstiness, in plain terms
Burstiness is the variance in sentence length and complexity across a passage — the pattern where a person writes a twenty-eight-word sentence loaded with subordinate clauses and follows it with a four-word sentence that just states a fact. High burstiness is a hallmark of natural human writing, and low burstiness — sentence after sentence landing in roughly the same length band — is one of the patterns generated prose shows. Worth being precise about something here: this site reports a Naturalness Signal by the same name, computed the same way, on the text you paste in. That is a coincidence of vocabulary, not evidence of any relationship between the two products — burstiness is a term of art used by researchers and multiple tools independently, and our using it does not mean we are measuring what GPTZero measures, still less that we know what conclusion GPTZero would draw from a given number.
What a rewrite changes, and what it does not
A rewrite through this engine varies the phrasing and the sentence lengths, and both are mechanical consequences of what it does: it substitutes less-predictable wording for stock vocabulary, and it splits long, uniform sentences at natural clause boundaries. Be clear about which of those we can actually see, though. This site computes three Naturalness Signals — Cliché Density, Burstiness and Lexical Richness — and it computes no perplexity at all, so we can show you a Burstiness figure before and after a Run, and we cannot show you a perplexity figure for either. The link between a Run and anyone's perplexity score is reasoning about the mechanism, not a number we have measured, and on a page about not overclaiming it would be poor form to blur the two. What a Run does not change is anything about the substance of the writing: whether the argument is yours, whether a claim is properly sourced, or what any particular classifier will conclude about the result on any given day. Classifiers are retrained on new data, thresholds get adjusted, and a tool tuned against last month's version of GPTZero has no claim on what next month's version decides.
Why we will not tell you what it will say
We do not run GPTZero against your text, we have no arrangement with GPTZero, and we do not test our output against their classifier before or after a Run. A product that promised you a specific verdict from a system it does not operate would be promising something it has no way to observe, and we are not going to do that. It is also worth saying plainly that false positives on genuinely human writing are documented and discussed by researchers and by students who have been on the receiving end of one — that is somebody's real, stressful problem, and it deserves to be taken seriously rather than waved away by either side of this argument. The honest posture toward any classifier's confident-looking number, including GPTZero's and including the Naturalness Signals on this site, is to treat it as one measurement among several rather than as a fact about whether a person wrote something.
About Rewriting for GPTZero
Are you affiliated with GPTZero?
No. We have no partnership, licensing arrangement or data-sharing relationship with GPTZero, and no GPTZero logo or trademark appears anywhere on this page or this site. The name is used here only to describe, in plain terms, what their published methodology measures, in the same way a review might describe a competitor's product without endorsement from it.
Will a rewrite change what GPTZero reports?
We cannot tell you that, and we are not going to guess. We do not run GPTZero against your text, so we have no basis for predicting what their specific classifier will report on any given passage, before or after a rewrite. What we can tell you is which of their documented measures a rewrite mechanically affects, which is the substance of the section above.
Your Burstiness figure and theirs — are they the same number?
No, and there is no reason to expect they would be. We compute our Burstiness figure from your text using our own method, and it describes sentence-length variance in what you pasted in — nothing more. GPTZero's methodology is theirs, not published in the detail ours is, and not something we have access to or attempt to reproduce. The shared name reflects a shared term of art in the field, not a shared calculation.
My own writing was flagged. What can I do?
First, keep whatever evidence you have that you wrote it: drafts, version history, an outline, notes, timestamps in whatever tool you wrote it in. That evidence is what actually resolves a false-positive dispute with an instructor or an institution, not a rewriting tool and not a different classifier's verdict instead. A rewrite here can change how your phrasing reads, but it cannot manufacture a history you did not keep, and it is not a substitute for the conversation you need to have with the record you do have.
What makes Un-AI Text a transparent GPTZero alternative for rewriting?
Unlike closed-box humanizers that claim impossible detection immunity, Un-AI Text gives you full editorial control with side-by-side sentence diffs, editable output panes, and transparent Naturalness Signals. You review and approve every restructured sentence before keeping it.