Paste the Claude draft into the box above and press Humanize. Every sentence gets rewritten in one pass, up to 12,000 characters or roughly 2,000 words per run, at one credit per fifty words. Ten credits a day are free per network, with no account, no card and no email address to verify. Run the checker on the result for a score from 0 to 100 and one of three verdicts: reads as AI-generated, borderline, or reads as human-written. Claude output has its own habits, and most of them are hedges, asides and address to the reader rather than the stock vocabulary people watch for. This page names them and gives the edit for each. This site is not affiliated with Anthropic.
Claude writes like someone talking you through a problem. That is pleasant to read, and it leaves a residue: qualifiers on claims that do not need them, asides that announce a point before making it, and a reluctance to land on one side of anything. The stock machine vocabulary turns up less than you would expect. The posture is the giveaway.
The in-browser tell registry highlights the entries below in place, with the reason for each in one line. It is free, it is unlimited, and it spends no credits. Not every flag is wrong for your context, so keep the ones you meant.
A chat reply is written to a person who just asked a question, and the shape of that reply travels into your document intact. Read the first line and the last line of a pasted draft first, because that is where the address to the reader sits.
Claude has a cadence. A long sentence carrying two or three clauses, then a short one for emphasis. Long, short. Long, short. Read a page aloud and you hear a metronome even after every flagged phrase is gone, because the pattern sits in the sentence lengths rather than the words.
Break the alternation rather than shortening everything. Put two long sentences next to each other. Let a paragraph close on the long one. Where a short sentence is doing real work, keep that one and rewrite the other three.
The registry scores sentence-length variation directly, so a flat distribution shows up as its own flag. It also flags an em-dash flood, three or more in a passage once they pass roughly 1.6 per hundred words, and three sentences in a row opening on the same word, which here is almost always "The" or "It".
These were written for this page rather than copied from a session. The move is the same in each. Delete the frame, then write the thing the frame was protecting you from writing.
The rewrite takes 12,000 characters per run, about 2,000 words, with a minimum of 25 words. A blog post, a cover letter or a standard essay clears in one pass. Anything longer goes through in parts, split at your own section breaks so each run sees a whole idea. The word counter shows both character totals, and the with-spaces figure is the one the limit measures.
The checker is separate. It needs at least 65 words and costs one credit per run whatever the length, so there is no reason to skip the baseline reading. The rewrite costs one credit per fifty words, so ten free credits a day covers one check plus roughly 450 words of rewriting.
The tell registry, the proofreader, the word counter, the readability checker, the AI vocabulary checker, the plagiarism and originality check and the signed drafting record all run in your browser, free and unlimited, and none of them spends a credit.
Order matters more than effort. Running the humanizer blind works, but you learn nothing about the habits you will repeat next week.
A score measures how machine-made a passage reads. It does not identify a model and it does not identify an author, because the only evidence in a file of prose is the prose. Careful, formal, evenly paced human writing collects the same flags.
Liang and colleagues at Stanford HAI found in 2023 that 61% of TOEFL essays written by non-native English speakers were falsely flagged as AI across seven detectors. A narrow idiom range and a formal register look, to a classifier, like generated text. There is more at why detectors flag human writing.
If you expect to be questioned, the signed drafting record logs the session as you write: timestamps, how much you cut, and whether a large block arrived in one paste. It exports a signed file anyone can check at verify a report without an account and without contacting us. If an accusation has landed, you wrote it and were accused covers what to do next.
Rewriting a draft you were permitted to generate is ordinary editing. Where AI is not permitted, that rule covers the draft however it is edited afterwards. Read the policy that applies to you.
We make no claim about what any other detector will report. What happens here is concrete: every sentence gets rewritten in one pass, and you get a score with a verdict before and after, so you can read the change yourself.
Ours does not, and it is not built to. It returns a number from 0 to 100 and one of three verdicts about how machine-made the writing reads, and it never names a model. We do not speak for what other tools report.
Yes. The registry matches on phrasing, not on where the text came from, so the same entries fire on output from a chat window, an API call or a writing assistant. The hedges and the long-short cadence travel with the prose.
12,000 characters per run, roughly 2,000 words, with a 25-word minimum, at one credit per fifty words. The checker needs 65 words and costs one credit per run at any length. Ten credits a day are free per network, with no login, no card and no email address to verify.
That it reads as machine-made, which is not the same as being machine-made. These tells are common in careful formal writing and in anything edited toward a house style. Run the free tell registry to see which phrases are driving the number.
It depends on the rules you are working under. Rewriting a draft you were permitted to generate is ordinary editing. Where AI is not permitted in submitted work, that rule covers the draft however it is edited.