Two different questions, not one
The first question is statistical: does this writing resemble the patterns Turnitin’s indicator, or any style-based tool, associates with AI-generated text. That’s answerable, at least as an estimate, with tools like Turnitin’s own indicator or NeverPrompted’s style check.
The second question is a policy question, and no detector answers it at all: regardless of what a score says, did what you actually did, using AI to brainstorm, to draft, to edit, comply with what your institution’s academic integrity policy permits? Some policies allow AI for brainstorming but not drafting. Some require you to disclose any use. Some ban it outright. That’s a decision about honesty and process, not a percentage.
A low or zero AI-likelihood score doesn’t touch the second question at all, and treating the two as the same thing is exactly how tools marketed as “humanisers” mislead people: they sell an answer to the first question as though it settles the second.
What Turnitin’s indicator actually measures, and its limits
Turnitin’s AI indicator is a percentage estimate based on how closely a document’s style resembles patterns found in training examples of AI-generated text. Turnitin itself has published that this comes with real false-positive and false-negative rates, it’s a statistical estimate, not a certain read of who wrote something.
Because it’s a resemblance-based classifier, running text through a paraphrasing or “humaniser” tool can shift the score, sometimes considerably, by changing surface style. But that’s a style change. It says nothing about whether the underlying thinking, research, or drafting was actually yours. A low score produced this way is not evidence that you did the work.
What actually holds up
Genuine authorship is the thing that protects you, not a score in either direction: understanding what you wrote well enough to explain and defend every sentence, and disclosing AI use exactly as your policy requires, if it requires it at all.
If you used AI for a first draft, the honest path is rewriting it properly in your own words and understanding first, then separately checking your institution’s stated policy on AI assistance and disclosing accordingly. Treat those as two genuinely separate steps, not one.
No tool, including NeverPrompted’s, can promise a specific outcome against any detector, because these classifiers keep changing and none of them are testing a mark the writer controls. The only durable answer is writing, or genuinely rewriting, the work yourself and being straightforward about how you used AI along the way.
Wherever this page describes a result: a detected mark is not proof of authorship, and an absent mark is not proof of human authorship. NeverPrompted's on-device rewrite can reduce detectable evidence but cannot guarantee defeating a vendor's undisclosed watermark, on any tier.