Plainly

Paste writing. See what reads as machine-made, and why.

Your text

0 words · Checked in your browser. No account, no upload, no history.

Findings

craft Is it good writing? Evidence, specifics, structure.
tells How loudly does it carry machine signatures?

    Findings appear here as you type.

    What this does, and what it will not do

    It finds; you rewrite

    There is no “fix it” button, on purpose. Every finding cites a line and says what the problem is. The judgment about whether it matters is yours, and a rule you can explain away in one sentence should be explained away.

    Two scores, never one

    Craft and tells are separate because they pull apart. Lowering tells by wrecking craft gives you garbled text that beats a detector and reads terribly. The headline is the lower of the two, so you cannot buy a good score on one axis alone.

    Never invent a fact to raise a score

    The evidence rule wants a real name, number or date. If you do not have one, cut the abstract sentence — do not fabricate a statistic, a source or a testimonial. That is worse than the finding, and in the case of a testimonial it is illegal.

    Nothing is uploaded

    The whole checker runs in this page. Your text is never sent anywhere, never stored, and never seen by us — there is no server to send it to. Close the tab and it is gone.

    Where the rules come from, and what they are worth

    The rules are ported from ucd-voice, which gates each one by measurement rather than opinion. Against a genre-matched human corpus — long-form journalism published well before ChatGPT, set against machine prose written to a bare brief — most of the popular “AI word” lists ran backwards: humans used the marker vocabulary more than the machines did, and the “not just X, it’s Y” construction did not separate the two at all.

    What did separate them was structural: abstract-noun density above all, then compound density and repeated openings. So the vocabulary rules here score craft, not tells — they are still worth cutting, they are simply not evidence of who wrote it.

    Limits. These thresholds were calibrated on one house’s registers. A score is a statement about fit to those registers, not a universal verdict, and a page can score 100 while still being the wrong thing to say. Nothing here can tell you whether a machine wrote something; it tells you which habits the text has.