Paste the Grok draft into the box below and press humanize, and the rewriter goes through every sentence in one pass, so the pivots, three-item lists and stock vocabulary that chat output produces come out together. One run takes 12,000 characters, about 2,000 words, with a 25-word minimum, and costs one credit per fifty words. The free tier is ten credits a day per network, with no account, no card and no email address to verify. Run the checker on what comes back for a score from 0 to 100 and one of three verdicts: reads as AI-generated, borderline, or reads as human-written. The checker needs at least 65 words and costs one credit per run whatever the length. Grok output is the interesting case, because the casual voice hides these constructions from the writer without removing them. The jokes sit on top of the same negation pivots, rhetorical questions and consultant vocabulary that formal chat output produces, and the rhythm work sits in opener repetition and aside density rather than sentence length. This page names the habits, shows why the loose voice still reads as machine-made, and gives the rewrite for each, and it ends with the honest version of what a score can and cannot tell you about the result.
The humanizer takes 12,000 characters per run, roughly 2,000 words. A blog post or a standard essay clears in one pass. A 5,000-word paper takes three runs, so split it at section breaks rather than mid-argument, and the humanize 5,000 words page walks through that split. The minimum is 25 words and the cost is one credit per 50 words.
The checker is separate and costs one credit per run whatever the length. It needs 65 words and returns a 0 to 100 score plus one of three verdicts: reads as AI-generated, borderline, or reads as human-written. Run it before and after a rewrite to see which paragraph was carrying the score.
Several tools alongside it are free and unlimited and spend no credits. There is a proofreader, a word counter, a readability checker, an AI vocabulary checker, a plagiarism and originality check, and a signed drafting record, and they all run in your browser. The full registry, 44 tells in six groups, is written out at AI writing tells.
A Grok draft rarely gets caught on one exotic word. It gets caught on a cluster of ordinary constructions arriving together in a short span, and the conversational packaging does not hide them. The registry sorts its entries into six groups: structure, openers, transitions, lexicon, hedges and cadence. Each item below is fixable in a single edit.
These are the ones you will hit first in chat output written in a casual voice:
Swap every flagged word and a draft can still read as machine-made, because part of the giveaway is the shape of the sentences rather than the vocabulary. Grok output is an interesting case here. It mixes a four-word punch with a long winding sentence, which answers one rhythm entry and leaves two untouched.
The first is sentence-length variation. Human drafts vary their lengths, and a passage where every sentence runs about the same length gets flagged for uniform rhythm. Alternating short and long usually clears that on its own, so on a draft in this register the rhythm work is mostly in the other two entries.
Those two are opener repetition and punctuation density. Three or more sentences in a row opening on the same word gets flagged, and conversational output starts sentence after sentence on And, But and So. The em dash entry fires when a passage carries three or more of them, and a voice built on asides reaches that quickly. Fix both by starting one sentence on its object and promoting two asides into sentences of their own.
Every fix below follows one of four rules. Keep the positive half of a pivot. Cut the opener and check whether the paragraph still stands, because it almost always does. Replace a rating adjective with the thing at stake. Put a specific where a quantity word sat.
Read them as pairs. The after version keeps the claim and drops the packaging around it. If your rewrite still sounds like the before version, the packaging survived, so cut it again.
Check first, so you have a baseline number to beat. Then read your draft against the registry at AI writing tells, which highlights each phrase in place with the reason it fired. Some flagged phrases are right in your context, and you should keep those. Then humanize, re-check, and read the result aloud once.
Register is the second pass and the one people skip. A draft written to sound good in a feed and a draft written for an assignment are different documents. Chat output in this style leans on second-person address, jokes that carry a paragraph and asides to the reader. Read the result once as the person who will receive it: an editor, a grader, a hiring manager. Anything that only worked as a post comes out.
A rewrite does not check facts. Any name, date, number or citation that came out of the chat window still needs checking against a source, and models produce citations that look right and do not exist. One specific that only you know also does more for a paragraph than any amount of rephrasing, so add yours while you are in there.
On academic use the site takes one line. 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.
A detection score measures how machine-made a piece of text reads. It does not measure who wrote it, because the only evidence in the file is the prose itself. Careful formal human writing collects the same tells, and so does anything edited toward a house style. The score page explains the number in detail.
That matters most for writers who get flagged without ever opening a chat window. 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 formal register and a narrow idiom range look, to a classifier, like generated text, and the ESL writers page covers that situation directly.
If you wrote the piece yourself and expect to be questioned, the signed drafting record is free and unlimited. It records the session as you write, including timestamps, how much you cut, and whether anything large arrived in a single paste. It exports a signed file, ECDSA P-256, that anyone can check without an account and without contacting us. If an accusation has landed, the falsely accused page covers what to do next.
We make no claim about what any other detector will output. What the humanizer does is rewrite every sentence and take out the constructions in our registry of 44 tells. Run our checker on the result and read the score yourself.
Because a score reads structure, not tone. The jokes sit on top of the same negation pivots, three-item lists, rhetorical question pivots and consultant vocabulary that formal chat output produces. A conversational draft also runs into the opener repetition and em dash density entries, because sentences starting on And, But and So, and asides set off with dashes, are part of that voice.
12,000 characters per run, about 2,000 words. For anything longer, split at a section break and run the parts separately so each run works on one argument instead of a whole paper.
No. The free tier is 10 credits a day per network with no login, no card and no email address to verify. The humanizer costs 1 credit per 50 words and the checker costs 1 credit per run at any length.
The humanizer needs 25 words and the checker needs 65. Below that there is not enough text to read.