The Judge
Paste a draft. This counts what is measurably in it — how your sentences vary, and how often four grammatical structures appear that show up far more in machine-written prose than in human prose. The measuring runs in this tab: your draft is never sent anywhere, and there is no endpoint here that could receive it. (The site counts a pageview, like every page. It does not see a word you paste.)
It will not tell you whether a machine wrote your draft. Read the line it will not draw for why not, and what it would take.
—
Shape
Paste something above.
Deterministic passes
Paste something above.
The tells, measured
Paste something above.
How to read this
Every rate above is counted from the text you pasted, exactly. None of them is an estimate and none of them is a verdict. A rate on its own is a fact about your prose and nothing more.
The four structures come from Reinhart et al., who measured how often each appears in machine-generated prose against human-authored prose. Their ratios are reported next to each one. A ratio is not a threshold: it only means something when it is applied to a human baseline measured from real writing. Applied to nothing, it is a number with no denominator.
The line it will not draw
One sample cannot tell you whether you sound like a machine. It can only tell you what is in it. To say a rate is unusual, something has to be usual first — and the only honest reference is a body of your own earlier writing, because human prose varies so widely that a general baseline flags real people as machines.
That is not hypothetical. Drawn against a naive universal line, this exact measurement called one of my own hand-written essays machine-written. The fix was not a better threshold. It was refusing to draw the line at all until there is enough of your work to place it.
OneWriter is the writing app this measuring engine came out of — literally this file, running on your machine instead of in this tab. It keeps your manuscripts, builds the baseline from your own back catalogue, and then the bands above stop being blank. It exports EPUB, DOCX and print-ready PDF, and it will not write a sentence for you. $29, one time. You get the source.
GPL-3. Linux and macOS, Node 24+. If you would rather build it yourself, that is what the licence is for.
Frequently asked questions
Can a tool tell whether my writing sounds like AI?
Not from one draft. A single sample has nothing to be unusual against, so anything that claims a verdict from it is guessing. The Judge counts what is measurably in your text and stops there. A real comparison needs a stack of your own earlier writing to hold the new draft up to.
What does The Judge measure?
Two things. How much your sentence lengths vary, and how often four grammatical structures appear that Reinhart et al. found far more in machine prose than in human prose: participial tails, nominalizations, "that" clauses, and phrasal coordination. Every rate is counted from your text exactly, never estimated.
Does it upload my draft, or need an account?
Neither. The measuring runs in your browser tab. There is no endpoint on this page that could receive your draft, and nothing to sign up for. The site counts a pageview, the way every page does. It never sees a word you paste.
Why do AI detectors flag human writing as machine-written?
Because they draw one universal line, and human prose varies too widely to sit under it. This same measurement, drawn against a naive universal baseline, once called one of my own hand-written essays machine-written. The honest answer is to refuse the line until there is enough of your work to place it.
How do I keep my own voice when I edit with AI?
Measure against yourself, not a general standard. OneWriter builds the baseline from your own back catalogue, so the comparison this tool leaves blank fills in against how you actually write. It keeps your manuscripts and will not write a sentence for you. $29 once, and you get the source.