Menu
Start

Free tool

Your draft's fingerprint, in 24 numbers.

Paste a post. We measure how it's built, from sentence length to how often it repeats the same beat, and put each number beside the range where good published writers sit. Two runs a day are free without an account. We don't keep your text.

150 to 5,000 words, Markdown is fine0 words

Longer texts give steadier results. For the best read, paste about 2,500 words of one original text. Pasting the same text twice changes nothing.

How it works

Counted, then compared.

We split your draft into sentences, read each one's grammar, and count. Headings, lists, quotes and tables stay out. Every count is then compared with the same count on about 6,000 words from each writer in our reference set: essayists, authors and company blogs people read on purpose.

GreenWithin one spread of the writers' middle. You write this part the way they do.
AmberOne to two spreads away. Worth a look.
RedMore than two spreads away, too much or too little. Where a draft has too much of something, we show the sentences that counted.

The score out of 100

Each measure is free inside the green. Past it, the cost grows with the square of the distance for the first three spreads, then in a straight line, so one habit far out of range can sink a post on its own: a draft that is green everywhere but ten spreads out on one measure scores 78. On Funk, writing less than the writers never costs anything. A group's cost is the average of its measures, the five groups weigh the same, and the score falls away from 100 as the cost climbs. The writers in our reference set score between 79 and 99, each measured against the others. We call this scale W10V4.

For each measure we drop the single writer farthest from the rest, then take the others' average as the middle and their standard deviation as the spread. No measure is a target. Good writers differ, and a green fingerprint is no promise of a good post.

What a prompt changes

Claude with no style instructions, against our writer.

We gave Claude one brief from our own blog, a B2B content plan for founders, with no instructions on style, and scored the drafts. Then we scored our writer's drafts of the same brief, four more briefs after W10's rules went into its instructions, and one long draft on a fifth brief before and after the writer reviewed its own sentences against the fingerprint.

Claude, no style instructionsOur writer, 26 OctoberOur writer, with W10's rulesOur writer, after its self-review
Score64226688
Drafts4641
Words each1,4001,3001,1002,700
"You", everywhereLeft alone, Claude addressed the reader 4 to 5 times per 100 words. The writers in our reference set do it about once. Our writer's instructions brought it down to about 2, and W10's rules into range.
A fix that cost somethingOur October instructions evened out the voice and made the specifics worse: in about two thirds of main clauses the subject was a type ("a founder", "most teams"), against nearly half for Claude alone and a fifth for the writers. They also chopped the sentences short. W10V4 bills a habit that far out heavily, which is why that column scores below Claude alone. W10's rules brought the writer back level.
What moved itRules in the instructions got the writer level with Claude alone. The self-review did the rest: the fingerprint marks the sentences where a type stands in for a person or a thing thinks, the writer rewrites only those, and the same draft went from 67 to 88. Specifics are still where it stays furthest from the writers.

Scored with W10V4. The first two columns are drafts of one brief; the third is four other briefs for the same blog, so it compares the same writer on different subjects; the fourth is one 2,700-word draft of a fifth brief, written on 30 October 2026, scored after the self-review (67 before it). The self-review is in testing and not yet part of our writer. Drafts of the same brief vary by about 10 points, more at 1,300 words than at 2,500. "No style instructions" means Claude got the brief and nothing else. "26 October" is our writer's instructions as they were that day, before W10's rules.

The 24 measures

Five groups, twenty-four counts.

MeasureWhat it counts
Rhythm How long the sentences run and how much they vary. How it works
Sentence lengthAverage words per sentence. Good prose sits near 19.
Length swingHow much lengths vary. A long sentence after two short ones keeps a reader awake.
Short sentencesSentences of 8 words or fewer. A few land a point; many make the text bark.
Long sentencesSentences of 25 words or more. Writers people enjoy use plenty of them.
Drumbeat stretchesSix sentences in a row, all short and alike in length. Reads like a metronome.
Flow How one sentence leads into the next. How it works
OpenersSentences that start with something before the subject: "In 2019,", "When the call ended,".
ConnectorsBut, because, which, unless, then. The words that tie one thought to the next.
Cause wordsSentences that say why: because, so that, which means.
Flat stretchesSix sentences in a row that all open on their subject.
Voice Who the text talks to, and whether it tells what happened. How it works
YouYou and your, per 100 words. Some is warm; a lot sounds like a sales page.
Past tenseSentences about something that happened. Stories and evidence live here.
Specifics Whether it talks about real people and things, or about types. How it works
Type subjectsClauses whose subject is a type: "every founder", "most teams", "buyers".
Made-up scenesA type doing something at an invented time: "A founder opens it on Monday."
Type actionsA type acting in the present: "Buyers skim."
"The" actionsThe same with "the": "The founder opens the dashboard."
Type statesA type in a state: "A buyer needs proof."
Funk The tics readers have learned to spot in machine-made prose. Only too much counts. How it works
DashesEm dashes per 100 sentences. Writers use about nine; a dash in every other sentence is a tell.
"Quietly" wordsQuietly, silently, subtly, seamlessly. Adverbs of a manner nobody can check.
Not X, but YA claim set up only to be knocked down: "It's not a writing problem, it's a planning problem."
-ing tailsA sentence that trails off on a comma and a verb in -ing: "..., making it easy to share."
Do-and-done chainsAn instruction tied to its result with "and": "You say it three times and it sounds like AI."
Things with mindsA thing that speaks, knows, wants or agrees: "Your posts disagree."
Stops being X, becomes YA change told as a before-and-after: "It stops being a blog and becomes a sales team."
AnnouncersAnnouncing the point before making it: "The hard part is...", "Here's the thing".

API

Same scorer, from Claude or your own code.

This page calls a public API, and so can you. Send a draft, get the same 24 numbers in their five groups as JSON. The same limit applies: 20 runs a day.

curl -s https://app.propaganda.pub/w10/v1/score \
  -H 'Content-Type: application/json' \
  -d '{"text": "Your draft, 150 words or more..."}'

GET https://app.propaganda.pub/w10/v1/ describes the API.

Questions

Before you paste.

Is this an AI detector?
No. It can't tell who or what wrote a text. It measures how the text is built and shows where that differs from writers people read by choice.
Do you keep my text?
No. Your draft is scored in memory on our server and thrown away. We don't store it, log it or train on it. We count how many runs each visitor makes in a day, and that count resets at midnight UTC.
Why 20 runs a day?
Every run reads the grammar of every sentence on our own server. Twenty a day each keeps it free for everyone, with no account.
Which writers are in the reference set?
Essayists, authors and company blogs, about 6,000 words of published prose each. We show their numbers, never their text.
My draft is all green. Is it good?
It's built the way good prose tends to be built. Whether it's true, useful or interesting is a different question, and no count answers it.
What does the line at the bottom of a result mean?
The scorer's version and the reference set it used, for example "W10 1.1.0 · reference e4ee56d27c9a". Two results compare only when both match.

Propaganda

Our writer is briefed on these same measures.

Propaganda plans your company blog, drafts every post in your voice, checks it against what you know to be true, and waits for your review.

See how it works