Paste it. neepora rebuilds every sentence on the skeleton of real human writing, then runs GPTZero and shows you the result next to the output. Under 90 and your credits come back.
The three samples are the actual GPTZero 4.9b results from 2026-09-15. Each draft scored AI 100%; each rewrite scored Human 90%. See the full log
Same detector, same day, same account. Not the best runs picked out: the failed attempts are in the log too.
| Document | Words | Draft | After | Flagged sentences |
|---|---|---|---|---|
| Sample master's thesis, ch. 1–4 (regulation comparison) | 9,017 | AI 100% | Human 90% | 0 of first 10 shown |
| Sample master's thesis, ch. 5–7 (data, discussion) | 4,263 | AI 100% | Human 89% | rescan 88–89 |
| Language-learning essay (one page) | 622 | AI 100% · 21/23 | Human 90% | 0 of 35 |
| Stablecoin policy brief (one page) | 471 | AI 100% · 16/16 | Human 90% | 0 of 16 |
| Encyclopedic article on tea (one page) | 774 | AI 100% · 23/23 | Human 90% | 0 of 36 |
GPTZero Model 4.9b, paid account, 2026-09-15. Rescans vary by ±3 points.
Each genre borrows its sentence structure from a different kind of human writing: essayists for essays, official documents for theses, real correspondence for letters.

AI text is smooth, so it is predictable. Swapping words, cutting sentences and fixing punctuation left the verdict unchanged. Eleven registers, eleven AI 100% results.

A human text close to your subject, most of its words kept, only what the subject forces changed. No sentence of your draft survives. No invisible characters, no deliberate typos.

Long documents are processed in 600–1,000-word sections and rejoined. Under 90 we swap the donor and run once more; still under, the credits come back.
The borrowed structure shows. A thesis comes out sounding like an official document; an essay sounds like the essayist it borrowed from. A slightly awkward phrase may remain, and that is part of why it passes. Check your numbers and citations before you submit.
This is not built to break your school's rules. Read our principles

1 credit = 100 words. Packs, not subscriptions. Use them in exam season, leave them alone in summer; they are still there.
On our measured documents, yes: Human 89–90% on a 12,800-word thesis and 90% on three one-page tests. We publish every result, including the failures, on the results page. Under 90 your credits come back.
Yes. The facts, order and claims stay. What changes is the sentence structure, which is what detectors read. The style of the human donor text will show.
It depends on your rules. If your school or employer forbids AI drafting, do not use it. If AI drafting is allowed, neepora keeps a machine-sounding draft from being judged on its sound rather than its content.
No. Those break with the next detector update and are dishonest. We change sentence structure only.
Supported. Korean counts 200 characters as one credit. GPTZero's Korean verdicts are rougher than English, so the 90 guarantee applies to English documents.
About a minute per 1,000 words. A 20,000-word thesis is back in under half an hour with a per-chapter score.