QuillBot's AI detector: the AI-refined category and how to read human on it
QuillBot is best known as a paraphraser, which makes its AI detector an interesting case. It is known for a free checker that does not give one percentage but splits text into categories, typically along the lines of AI-generated, AI-refined and human-written, and it is aimed squarely at students. Humanize 360 takes a different view of the same problem: rather than sorting text into buckets, we show the eleven measurements underneath and let you change the ones that matter.
The category that catches most people is the middle one. Text that a person wrote and then polished with a paraphraser or grammar tool often lands in "AI-refined". That includes text polished by QuillBot itself. No tool can promise which bucket QuillBot will pick, because its model changes and its categories are its own. But the reasons text lands in the middle are measurable, and they are the same reasons every detector reacts to.
Three buckets, not one score
According to its public documentation, QuillBot's detector labels passages rather than scoring the whole document with a single number, and it is free to use up to a word limit. The labels are meant to distinguish text written by a model from text a human wrote and then had a model rewrite. In practice the boundary is statistical: the detector is looking at word predictability, sentence rhythm and phrase choice, the same things our breakdown reports as perplexity, burstiness, stock-phrase density and vocabulary spread.
The AI-refined trap
A paraphraser that runs on a language model produces language-model output. If you write a paragraph yourself and then push it through a model-based rewriter to make it "flow", you replace your slightly uneven human sentences with smooth, predictable ones. Grammar assistants that suggest sentence rewrites do the same in smaller steps. The result reads polished and lands in the middle category. Students find this confusing because they did write the essay. The detector is not disagreeing; it is noticing the polish.
How Humanize 360 handles the middle category
- Our engine is our own code, not a language model, so it does not add model-shaped smoothness. It measures your text and adjusts rhythm, openers and phrasing toward how people write.
- Paste 300 words of your own unpolished writing as a custom style profile. The engine measures your contraction rate, sentence variance and vocabulary spread and matches them.
- Use Light strength for text you wrote yourself. It is enough to lift variance and remove stock phrases without rewriting your argument.
- Use selective rephrase for the two or three sentences the breakdown singles out, and leave the rest alone.
- Check, humanize, check. If the breakdown reads human and the essay sounds like you aloud, stop.
Students and the disclosure question
Many institutions now allow AI tools for editing and grammar with disclosure, and prohibit them for drafting. Read your course policy before you use anything, including QuillBot. Where disclosure is required, disclose. Our Acceptable Use Policy at /acceptable-use rules out using Humanize 360 to breach academic rules, and /ethics explains why we think honest editing and dishonest submission are different things.
Common questions
Why does QuillBot say my own essay is AI-refined?
Most often because it was polished by a model-based tool, including QuillBot's paraphraser or a grammar assistant's rewrite suggestions. The polish is what the category detects. Our breakdown shows which signals changed.
Is Humanize 360 a paraphraser like QuillBot?
We have a free paraphraser at /paraphraser, but the engine is our own code, not a language model. It changes rhythm, openers and phrasing without adding model-shaped smoothness or sending your text anywhere.
Can you promise QuillBot will label my text human-written?
No. Its categories and model are its own and they change. We remove the statistical tells that all detectors react to and show you the evidence. That is the honest limit of what any tool can do.
Should I use Light or Heavy strength on my own writing?
Light. Your writing already has your voice; it just needs a few tells removed. Heavy is for text that a model drafted from scratch.