Undetectable AI vs Turnitin: a humanizer and a detector are not rivals

Because every field that would make a comparison is different in kind, not in degree. One is bought by a student for a few dollars a month and returns rewritten text; the other is bought by a university and returns a percentage the student never sees.

Turnitin's product page states that its checker reports how much of a submission is likely AI-generated, "including likely AI-paraphrased or bypassed content" (Turnitin, AI writing detection, read 9 November 2026). Its FAQ gives the detail, and the detail is where the limits sit.

Two of those rows settle most of the search. Turnitin says humanized text is in scope, and Turnitin publishes no rate for catching it. Anyone who tells you a humanizer passes or fails Turnitin at a stated rate is not quoting Turnitin. Turnitin's detector is not available to the public, so this site relies on Turnitin's published statements and peer-reviewed studies and reports no Turnitin result for any tool, this one included. The question of whether Turnitin detects humanized text has its own page.

Nothing measured, on the pages read for this article. Its pricing page states plans and the trial size, checked on 8 November 2026. Its detector page describes a checker that "shows the AI detection probability from major checkers, helping students and professionals understand whether their text could be flagged" (Undetectable AI, AI detector, read 9 November 2026). That is a prediction about other detectors, made by the company that sells the humanizer on the same page, and it is not a Turnitin reading because no company outside Turnitin can produce one.

The one peer-reviewed test that included Turnitin is from 2023. Weber-Wulff and colleagues ran 54 documents through 14 detectors, Turnitin among them, and found that Turnitin was the only tool to classify every unmodified AI document correctly, that no tool classified every tool-paraphrased AI document correctly, and that average accuracy on tool-paraphrased text across all 14 was 26 percent (Weber-Wulff et al. 2023, International Journal for Educational Integrity, read 9 November 2026). Those are findings about 2023 models and one paraphrasing tool, QuillBot, not about Undetectable AI, and Turnitin has updated its model since, most recently in October 2026 by its own account.

No detector score should be the only evidence in a misconduct case. That sentence is Turnitin's position as much as this site's, and it cuts both ways: a clean report is not proof the work is yours, and a flagged report is not proof it is not.

Answer the question underneath the search, which is whether an AI draft, humanized or not, is allowed in the work being submitted. If the course bans AI tools, the detector's reach is beside the point: unacknowledged AI content in assessed work is misconduct under the Cambridge, Oxford and Harvard policies read on 9 November 2026, whatever Turnitin shows. If the course permits AI assistance, disclose it in the format your style guide requires, and then the Turnitin percentage becomes a number your instructor already expects.

The same logic applies to "Walter Writes AI vs Turnitin" and "Grubby AI vs Turnitin": a humanizer and an institutional detector are never a pair. The twelve humanizers are set beside Turnitin's statements, with no result for any of them, on the Turnitin evidence page, and the comparisons this site does publish, humanizer against humanizer on the same inputs, sit in the compare hub.