GPTinf review

GPTinf's cheapest plan, Lite, costs $4.99 a month for 5,000 humanizer words, read on its pricing page on 9 November 2026. This GPTinf review covers what the three plans buy, what the vendor states about its output, and where the tool sits in the full humanizer ranking under one published method.

What does GPTinf itself claim?

The pricing page describes a built-in Ultra Humanizer as the vendor's strongest engine and attaches a blanket claim about AI scores to every plan, with no input set, detector list or date on the page, read 9 November 2026 (GPTinf pricing page). The FAQ on the same page states that all plans are month to month, that a buyer can cancel or switch tiers in two clicks, and that a remaining word balance stays available until the period ends.

The home page adds figures of its own: an average AI score under 5 percent after humanization and a detector that cross-checks eight systems, read 9 November 2026. Both numbers are the vendor's measurements on inputs the vendor chose, and the page does not name the detectors behind the average or say when it was taken. The same page is clearer about mechanics: a Freeze control that locks chosen terms so the rewrite leaves them alone, eight writing modes, and a re-humanize control for selected passages.

What does the research say about paraphrased text?

GPTinf is a paraphraser, and the published research on paraphrasing explains why a rewrite moves detector readings and why that is not the whole story. Weber-Wulff and colleagues tested 14 AI text detectors and found every one below 80 percent accuracy, with accuracy falling further on paraphrased and machine-translated text (Weber-Wulff et al. 2023, International Journal for Educational Integrity, read 9 November 2026). Krishna and colleagues showed that a paraphrase model 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, read 9 November 2026).

Detector makers answer in kind. GPTZero states that its model shields against common evasion methods, paraphrasing among them, and that it keeps its false positive rate at no more than 1 percent when evaluating AI against human text (GPTZero technology page, read 9 November 2026). A humanizer and a detector are moving targets for each other, and a reading is true for its date.

What none of those pages measure is whether the rewrite kept the writer's claims. That is why the method on this site weights meaning kept at 35 points of 100, errors added at 20, detector pass rate at 30 and limits and access at 15. No detector score should be the only evidence in a misconduct case.

What this GPTinf review draws on

Everything above comes from the vendor's pricing and home pages and the two studies linked. The run figures, when published, sit on the method page and in this review, with the raw outputs free to download.

What does GPTinf change in your text?

The home page describes the humanizer as trained on human text and says it rebuilds rhythm rather than swapping words, contrasting its own engine with what it calls a feather-light rewrite that changes vocabulary and leaves sentence structure alone, read 9 November 2026. The eight modes are named as writing modes rather than strength levels, and the Freeze control lets a user lock terms, which matters for anyone whose text carries names, figures or quotations.

The site's protocol runs each tool on its default mode and, where a stronger documented mode exists, that mode as a second row. GPTinf's Ultra Humanizer is described as always on, so a published run will record which of the eight modes was the default and whether any mode is documented as stronger. The reader-side costs the method rates are claims dropped, claims invented, grammar errors introduced per 100 words, and readability change. Those figures appear here once a run is published.