GPTZero humanizer: what GPTZero publishes and the tools listed on one set of fields

The table below lists eight tools in order of the largest request a cheapest paid plan or free tier accepts, largest first, because a GPTZero scan is run on a document and a tool that takes the document whole produces one rewrite rather than a stitched set. No column is a GPTZero reading. The GPTZero column of the published run will be added to this table when the first run publishes. Every figure comes from the vendor's own pricing or product page, checked on 8 November 2026.

None of those lines mentions GPTZero's reading, because none can yet.

GPTZero's technology page, read 9 November 2026, describes an end-to-end deep learning model trained on text from the web, education and a range of language models, with a sentence-by-sentence classifier that returns a probability and a confidence for each sentence (GPTZero, technology, read 9 November 2026). The same page names a Paraphraser Shield that it says defends against common methods used against detectors, "such as paraphrasing and homoglyph attacks", and a three-way document label: written entirely by a human, entirely by an AI, or a mix.

Three things about that table matter for a humanizer search. First, every figure is GPTZero's own, measured on data GPTZero chose; no independent body published them. Second, GPTZero publishes no rate for paraphrased or humanized AI text specifically, only the statement that its shield is designed to catch it. Third, GPTZero's confidence categories mean a reading is not one number: a document can be labelled AI with low confidence, and GPTZero says its calibration is biased toward false negatives over false positives.

What does GPTZero tell schools to do with a score?

GPTZero's own advice to educators is the most useful sentence on this page for anyone who was flagged. Its educators page says that like all detectors GPTZero "isn't perfect", that short or highly edited passages make detection harder, and that a scan should be "a conversation starter" rather than a verdict; it tells teachers to ask students to show their process through edit history or in-person assessment and to check patterns across several assignments (GPTZero, for educators, read 9 November 2026).

No detector score should be the only evidence in a misconduct case. That line comes from GPTZero's own guidance as much as from this site, and it cuts both ways: a reading that drops after a humanizer pass has not made the writing yours, and a reading that rises on your own prose has not proved anything by itself.

Why does this page rank nothing by a GPTZero reading yet?

Because a reading is only meaningful with a date, a version and a method attached, and this site has not published a run. The published method scans every output on five detectors, GPTZero among them, records GPTZero's own label and confidence on the run date with a screenshot, and counts "pass" only where GPTZero's own label is human. Each input is also scanned before humanizing, so the change a tool makes is visible rather than inferred.

The peer-reviewed record shows why the baseline matters. Krishna and colleagues reported that their DIPPER paraphraser dropped DetectGPT's detection rate from 70.3 percent to 4.6 percent at a 1 percent false positive rate (Krishna et al. 2023, arXiv 2303.13408, read 9 November 2026). That study did not test GPTZero's current model, and a 2023 result on one detector says nothing about a 2026 reading on another. What detectors measure in general has its own explainer, what AI detectors measure.