Rewrite a ChatGPT Draft So It Reads Like You Wrote It
A chat assistant writes competent, evenly-weighted prose that sounds like every other chat assistant. This page runs the same engine as the rest of the site, with Standard Mode selected, over whatever you paste in. Up to 250 words without an account.
Your AI draft
Humanized output
Detector check includedRun the humanizer to see detector results.
Detector verification status
What makes a chat draft recognisable
Three things, and none of them is grammar. The first is vocabulary: a narrow band of words appears at many times its natural rate — delve, tapestry, pivotal, underscore, landscape, realm, multifaceted, robust, seamless — alongside connective scaffolding like furthermore, moreover and it is important to note that. The second is sentence rhythm. A model optimising for the likeliest next token produces sentences of similar length and similar clause structure, paragraph after paragraph, where a person writes a long one and then a short one without thinking about it. The third is the shape of the argument: open with a framing sentence, three parallel supporting clauses, close with a restatement. Each of those is fine once. Together, over four paragraphs, they read as a machine.
What a Run actually changes
A Run has two halves. The first is deterministic and happens locally: fifty-one phrase substitutions drawn from the vocabulary above, twenty-two contraction pairs, and sentence splitting wherever a sentence runs past the Mode's threshold and a natural clause boundary exists. Hedge or informal openers are inserted at the second sentence of the passage depending on the Mode. Because this half is deterministic, the same Draft in the same Mode with the same Seed produces the same result every time, and re-running with a new Seed gives you a genuinely different set of choices rather than a reshuffle. The second half is a model pass that smooths what the first half broke, and it streams, so you read the Rewrite as it arrives rather than watching a spinner.
Why every change is highlighted
Rewriting tools that hand back a finished block of text ask you to trust them about your own argument, which is the one thing you cannot delegate. So the changes are computed on our side rather than requested from the model: both texts are segmented into sentences, aligned, and diffed, and every span that differs is marked in the output pane. You can read each one, keep it, or put your own wording back. This matters most in the places a rewrite is most likely to go wrong — a hedged claim that becomes a flat assertion, a technical term swapped for a near-synonym that means something slightly different in your field.
What it will not do
It will not do your thinking. A Run improves the phrasing of an argument that already exists; it cannot check a fact, find a source, or notice that your second paragraph contradicts your fourth. It will not make an unattributed source attributed — if a passage needs a citation before a Run, it needs the same citation afterwards. And it makes no claim about what any classifier will say about the result, because we do not run those classifiers and would not publish the result if we did. What you get is three Naturalness Signals computed from your own text: how many stock phrases were in it, how much its sentence lengths vary, and how varied its vocabulary is, each reported for the Draft and for the Rewrite so you can see the movement.
About ChatGPT Rewriter
Will it change the meaning of what I wrote?
It works at the level of phrasing and sentence boundaries rather than claims, and every sentence it touches is highlighted so you can read the change. The place to watch is hedging: if your original said a result suggests something, check that the Rewrite has not made it show something. Reject any span where that happened and your own wording goes back.
How long a passage should I run at once?
Whole paragraphs rather than single sentences. Sentence-length variance is a property of a passage, not a line, so the engine needs several sentences to vary against each other. Without an account you get 250 words a Run, which is about two solid paragraphs — enough to judge whether you like what it does. Pro removes the cap so a full piece keeps its rhythm across paragraph boundaries.
Can I get a different version of the same paragraph?
Yes. A Re-run uses the same Draft and the same Mode with a new Seed, which changes which synonym is chosen at each substitution point and where sentences split. It is not a reshuffle of the same output — the deterministic half of the engine genuinely picks differently. Run it two or three times and keep the one you like.
Does it work on text from Claude or Gemini?
Yes, and on your own writing. Nothing in the engine is specific to one model's output: the phrase list, the contraction pairs and the sentence-splitting thresholds describe machine-typical prose in general, and all three major assistants converge on much the same register. It is also useful on human writing that has gone stiff, which is most first drafts.