Paste an AI draft above. Every phrase that reads as machine-made is highlighted, named and explained in plain language, so you can rewrite it in your own voice. The way to humanize AI text is to see what marks it as machine-made first, and that is what this tool puts in front of you.
Drop in the AI text you want to humanize, or type straight into the box. The check runs in your browser, so there is no upload, no progress bar and no queue. Scoring starts at 25 words.
You get a 0–100 detectability score and a risk band. It measures how machine-made your text reads, which is a different question from who wrote it. The tool keeps those two apart.
Every finding is highlighted where it occurs and comes with a plain-language reason. 44 patterns, each one nameable: the negation pivot, the tricolon, the dead transition, the flat rhythm.
Take a suggestion, write your own line, or press Humanize and have the model rewrite every sentence at once. Suggestions are measured against the rhythm and vocabulary already in your document, so the result reads like you rather than like a thesaurus went through it. The one-click rewrite runs on our servers; scoring and tells stay in this tab.
Six families of tell, 44 detectors in total. These counts are read out of the shipped registry at render time, so the number on this page is the number running in your tab.
The negation pivot is one of the strongest single predictors of machine text. Say the positive thing directly.
Scene-setting that describes no scene. Detectors weight it heavily because almost nothing else opens this way.
These connect nothing. If the next sentence follows, it follows without an usher.
Post-2023 generated text uses this verb at many times the human rate. It is close to a fingerprint.
Two hedges in a row cancel each other and read as risk-averse machine output.
Three or more sentences in a row opening on the same word is parallel structure the reader can hear.
One admissions paragraph, scored live in both states. Nothing here is a drawn-on annotation: the marks come out of the same engine as the box at the top of this page.
GPTZero, Turnitin, Copyleaks and the rest never receive your writing from us. The score on this page comes from 44 patterns we can name and show you, which is why every number arrives with the sentences that produced it.
The checker itself is JavaScript running on your machine: 41 phrase patterns and 3 whole-document statistics, no model and no network call. You do not have to take that on faith. Open your browser's network tab, paste a paragraph, and watch nothing leave.
It also means the checker works in airplane mode. Once this page has loaded, scoring, highlighting and explanations run with the network switched off — on a plane, in a tunnel, or on a campus connection you would rather not trust.
All four are running in the box at the top of this page. Detectors retrain their classifiers without notice, so this tool predicts no specific score on a named one — it shows you the sentences driving yours.
44 detectors — 41 phrase patterns plus 3 whole-document statistics — run on your own processor as you type. No upload, no queue, no account, and it keeps scoring with the network switched off.
Every finding is highlighted where it occurs, with the pattern named and a plain-language reason attached. You can see the exact clause moving your score and decide whether to keep it.
The editor reads the sentence lengths, rhythm and vocabulary already in your document and holds suggestions within that range. The result sounds like the rest of your writing.
GPTZero, Turnitin, Copyleaks and the rest never receive your draft from us. Your score comes from patterns computed in your tab, which is why every number arrives with the sentences behind it.
Students arrive here for two different reasons. Some wrote the essay themselves, got flagged by a course tool, and need to know which sentences caused it before a meeting with a professor. Others used AI where the syllabus permits it and want the finished text to read like their own work. Both start the same way: run the paragraph, read the tells, decide what to change.
If you have been accused →If you learned English from textbooks, you write the constructions textbooks teach, and those are exactly the even, correct, unsurprising sentences classifiers score as machine-made. Stanford's 2023 study found 61% of TOEFL essays by non-native speakers falsely flagged across seven detectors. This tool shows which habits carry that risk without asking you to write worse English.
Written for ESL writers →Plenty of companies now permit AI drafting and still publish copy readers bounce off. Marketing, support and localisation teams use the humanizer to strip machine cadence out of permitted drafts before anything ships: the stock openers, the tricolons, the transitions that connect nothing. The check runs in the browser, so client copy under NDA gets scored without being uploaded anywhere.
Team pricing →Yes. The detectability score, the inline tells, the explanations, the rewrites and the export cost nothing and need no account. There is no meter and no document limit. Paid plans are on the roadmap; everything described on this page is what runs today.
No. There is no sign-up wall, no email box before the tool and no card. There is no account system at all — the text box at the top of this page is the entire product, working right now.
No. A single document can run as long as you like, and long drafts score the same way short ones do, on your own processor. Nothing is metered.
Yes, for the checker. Once the page has loaded, scoring, highlighting and report signing all run with the network off. Turn on airplane mode and use it on a plane; that is a real test, not a figure of speech.
The tell registry is English. The patterns it looks for — the negation pivot, the stock opener, the idle transition, the flat rhythm — are properties of English generated text. The rhythm and burstiness measurements are language-agnostic, so a non-English draft still gets a cadence reading, but the named tells will not fire.
That depends on the rules you are working under. Rewriting an AI draft you were permitted to generate is ordinary editing, the same as tightening a paragraph of your own. Where generative AI is not permitted in submitted work, that rule covers the draft however it is edited. The policy that applies to you is the one to read.
Back to Humanize AI, or open the full editor.