Does an AI humanizer cause plagiarism? Three separate risks

Whether the words in your document match words in a stored document. Turnitin's product page describes its Similarity product as a tool to "prevent plagiarism" and describes its AI checker separately, as a tool that helps "identify when an AI writing tool such as ChatGPT, text spinners, or AI bypassers (also called humanizers) may have been used" (Turnitin, AI writing detection page, read 9 November 2026). Two tools, two questions.

A humanizer is a rewriter. It takes the words you give it and returns different words that say, if it works, the same thing. A text-matching check compares those returned words with its index. Newly generated text, whether from a chatbot or a rewriter, usually matches nothing, which is why the similarity score on rewritten AI text tends to be low. Weber-Wulff and colleagues report a 2023 study by Khalil and Er in which 50 ChatGPT essays were fed to two text-matching systems and 40 of them, 80 percent, were judged highly original at a similarity score of 20 percent or less (Weber-Wulff et al. 2023, Testing of detection tools for AI-generated text, read 9 November 2026).

Yes, and this is the risk that is easiest to overlook. If the text you paste in paraphrases a source without crediting it, the text you get back paraphrases the same source without crediting it, in different words. The MLA Style Center's rule is that you cite a generative AI tool "whenever you paraphrase, quote, or incorporate into your own work" content it created, and that you "acknowledge all functional uses of the tool (like editing your prose or translating words) in a note, your text, or another suitable location" (MLA Style Center, How do I cite generative AI in MLA style?, read 9 November 2026). Rewriting a borrowed idea does not discharge the first duty, and using a humanizer triggers the second.

Weber-Wulff and colleagues name the middle case. They describe a student who "obtains a text written by an AI and then quickly goes through it and makes some minor changes such as using synonyms", call it patchwriting after Howard's 1995 term, and note that their 14 detection tools identified such text only about half the time (Weber-Wulff et al. 2023, read 9 November 2026). Patchwriting with a source is the oldest form of plagiarism; patchwriting with a chatbot is the same habit pointed at a different input. A humanizer automates the "minor changes" step and nothing else.

The citation row deserves one more line. The error classes on this site include lost attribution, where a claim the input credited to a source comes back stated as the writer's own. A rewriter that drops "according to the 2022 census" from the front of a sentence has turned a cited fact into an uncited one without anyone deciding to plagiarise.

No. Whether a text is your own work is a question about who produced the ideas and the words, and a policy answers it, not a detector. Turnitin's page says its AI checker reports "how much of a student's submission is authentic, human-written content versus likely AI-generated from ChatGPT including likely AI-paraphrased or bypassed content" (Turnitin, read 9 November 2026). The vendor is telling institutions that it treats rewritten AI text as a category of AI text, not as human writing.

Krishna and colleagues showed in 2023 why rewriting moves detector readings at all: their paraphraser dropped one detector's detection rate from 70.3 percent to 4.6 percent at a 1 percent false positive rate, and the authors described the result as evading detectors "without appreciably modifying the input semantics" (Krishna et al. 2023, arXiv 2303.13408, read 9 November 2026). Read that phrase carefully. The text still said what the chatbot said. The detector just stopped noticing. If the question is authorship, the answer did not change.

What a detector reads about your text is covered on the Turnitin page, and the broader question of whether using a humanizer is cheating has its own guide. The short version for this page: the humanizer changes the evidence, not the act.

Three checks, in this order.

First, the sources. Put the input beside the output and confirm every citation, attribution and quotation mark survived the rewrite; the lost-attribution fault is common and silent. Any idea that came from a source needs a citation in the output whether or not it had one in the input.

Second, the disclosure. Read your institution's policy, then the assignment's own instructions, and disclose what the policy requires. Under MLA guidance that includes the humanizer itself as a functional use. A disclosure written before submission is a note; the same facts discovered afterwards are a case.

Third, the meaning. A rewrite that kept your citations and your disclosure can still have changed a claim. The site's guide to humanizers and paraphrasers explains what each kind of tool does to a sentence, and the reading rubric this site uses lists the twelve ways a rewrite goes wrong.