ChatGPT humanizer: 12 tools on the same fields, and what ChatGPT text carries

The order below is by per-request cap, largest first, because a ChatGPT answer arrives as one block and a tool that takes it whole produces one rewrite rather than a stitched set. "Names ChatGPT" records whether the vendor's own product page lists ChatGPT as a source of text it rewrites, read on 9 November 2026. Every plan figure comes from the vendor's pricing or product page, checked on 8 November 2026. No column is a detector reading. The published run's 24 inputs include text generated with ChatGPT, and the ChatGPT subset column will be added to this table when the first run publishes.

None of those lines mentions a detector reading, because none can yet.

How long is a ChatGPT answer, and which caps fit it?

The caps matter because a ChatGPT answer is one piece of text and the caps are hard limits at the point of pasting. The published method uses 24 AI-written inputs across six genres, with the school essay paragraph set at 350 to 500 words, so that range is the yardstick this site uses for a typical long answer. A 500-word answer fits Phrasly, Clever AI Humanizer, HIX Bypass, StealthGPT and GPTHuman in one request and WriteHuman's Basic plan in one request; on Walter Writes AI's Starter plan it fits in one request only when it stays under 750 words, and a 2,000-word ChatGPT draft is three passes there. QuillBot free at 125 words per input is four passes for that same 500-word answer, and BypassGPT free at 80 words per input is seven.

Caps from each vendor's pricing or product page, checked on 8 November 2026. Verva, Undetectable AI and Grammarly state no per-request cap on their pricing pages, so they do not appear in the "one request" column; a cap that is not stated is not a cap of zero, it is a cap the vendor has not published.

Splitting a draft is not free of cost. Each pass is a separate rewrite with its own word choices, so terms, names and the thread of an argument can drift between the pieces, and the person stitching them back together is doing editorial work the tool did not do. That is why the order of the first table favours the larger cap.

Does ChatGPT text carry a watermark?

In the EU, OpenAI says yes. OpenAI's developer forum announcement of 5 October 2026 states that API customers worldwide can opt in to text watermarking for select models, that an invisible watermark would be added to eligible ChatGPT and Codex text generated in the EU over the following weeks, that it stays off by default in the API and would not be a global default at launch, and that the method, called textGrain, adds a statistical signal through the model's word choices (OpenAI Developer Community, OpenAI's approach to EU text provenance rules, read 9 November 2026). The same announcement says the detector's access is initially limited to approved researchers and expert organisations, that short passages and later edits make it less reliable, and that a missing watermark does not prove a person wrote the text.

The mechanism is the one Kirchenbauer and colleagues published in 2023: before each word is generated, a randomised set of "green" tokens is chosen and softly promoted during sampling, and a statistical test later counts whether a passage has too many green tokens to be chance, without access to the model (Kirchenbauer et al. 2023, arXiv 2301.10226, read 9 November 2026). Google states that SynthID Text, deployed in the Gemini app, adjusts the probability scores a model assigns to candidate words in the same way (Google DeepMind, SynthID, read 9 November 2026). The longer explainer is the AI text watermarks guide.

Two consequences follow for anyone searching for a ChatGPT humanizer. First, a watermark is a count over word choices, so a rewrite that replaces most words weakens it and a light edit leaves most of it in place; how far any one tool goes is a question for a run. Second, OpenAI's detector is not public, so no humanizer vendor can show you a textGrain reading, and a vendor page that sells a "ChatGPT watermark remover" is making a claim no public instrument can check. Several of the twelve vendors sell such a feature, read on 9 November 2026; this page records only what OpenAI documents.

What do the detectors this site uses say about ChatGPT text?

GPTZero's technology page states that it incorporates research in detecting ChatGPT, GPT4 and Google Gemini output among other models, uses an end-to-end deep learning model with a sentence-by-sentence classifier, returns one of three document labels, human, AI or mixed, with a confidence category, and keeps its false positive rate at no more than 1 percent when evaluating AI versus human text (GPTZero, technology, read 9 November 2026). The same page names a Paraphraser Shield that it says defends against paraphrasing and homoglyph attacks. None of that identifies ChatGPT as the author of a passage; a classifier labels text, it does not name the model.

That distinction sits under every "ChatGPT humanizer" claim. A vendor that says its output "passes GPTZero" is reporting a label on one text on one day, and GPTZero states that its model is updated as new language models appear. The published method records GPTZero's own label and confidence on the run date with a screenshot, scans every input before any tool touches it so the change is visible, and does the same on four other detectors. No detector score should be the only evidence in a misconduct case.

What can you do without a tool?

Two of the options people search for alongside "ChatGPT humanizer" cost nothing and involve no vendor. One is a prompt pasted back into ChatGPT asking for a rewrite against a list of named tells; the ChatGPT humanizer prompt guide gives three in full and sets out what the studies say a prompt alone can change. The other is editing by hand. Neither comes with a per-request cap, and neither comes with a published figure for meaning kept, which is also true of every tool above. The tools with a free tier are listed on their own page, the free humanizer list.

Students should check their institution's AI policy before using any writing tool. A humanizer does not make undisclosed AI use acceptable, and this site does not help anyone misrepresent authorship.