Stylometry: measuring a writer's fingerprint in numbers
Stylometry is the measurement of writing style. Instead of asking whether a text is good, it asks how it is built: how long the sentences run, how often "the" and "of" and "but" appear, how many words are used only once, where the commas fall. Those habits are hard to fake and hard to suppress, which is why stylometry has been used for a century to settle disputed authorship. AI detection is stylometry pointed at a new question: not which person wrote this, but whether a person wrote it at all. The eleven signals on Humanize 360 are stylometric measures.
The best known early case is the study of the Federalist Papers, where statisticians in the 1960s used the frequency of small function words to attribute the disputed essays between Hamilton and Madison. The insight still holds: authors do not think about how often they write "upon" or "whilst", so those counts stay stable across topics. Modern tools add sentence rhythm, vocabulary richness and punctuation, and they can be applied to machine output just as well as to a person.
What stylometry measures
- Sentence length: average, spread and how it changes across a piece.
- Function words: articles, prepositions and conjunctions, the words that carry grammar rather than meaning.
- Vocabulary richness: how many distinct words appear relative to total length.
- Punctuation habits: commas, semicolons, dashes, parentheses and how often each appears.
- Word and paragraph shape: opener patterns, paragraph lengths, list structures.
From authorship to authorship-by-machine
A language model has a style in exactly the stylometric sense. It favours certain transition words, ends sentences near a certain length, reaches for a list of three, seldom contracts, and spreads vocabulary evenly rather than fixating on a pet word the way people do. Because these habits come from the model's arithmetic rather than a personality, they are consistent across topics and across users. That consistency is what a detector reads. Our breakdown labels the habits by name so you can see which ones your text shares.
Where stylometry stops
Style is not identity. Two people trained by the same school write alike; one person writes differently in a text message and a thesis; a careful editor flattens everyone. So a stylometric match, human or machine, is a probability and never a proof. Detectors built on it inherit that limit, which is why a score should prompt questions rather than conclusions, especially in a classroom.
Using it on your own writing
Paste 300 words of your own prose into a Humanize 360 style profile and the engine measures your contraction rate, sentence length and variance, and vocabulary spread. Those numbers become targets for the humanizer, so the output leans towards your habits rather than a generic average. It is stylometry used constructively: measure the fingerprint, then write towards it.
Common questions
Is stylometry the same as AI detection?
AI detection is one application of stylometry. Classic stylometry asks which author wrote a text; detection asks whether the author was a person. The measurements overlap heavily.
Can stylometry identify me from my writing?
With enough samples and a small pool of candidates, sometimes. Across the open internet, rarely. Our style profile measures a few numbers for your own use and does not attempt identification.
Which Humanize 360 signals are stylometric?
All eleven. Sentence-length variance, vocabulary spread, contraction rate, punctuation patterns and opener repetition are the most traditional; stock-phrase and tell-word measures are newer additions aimed at machine habits.
Can I change my stylometric fingerprint?
Slowly and only partly. Conscious edits shift sentence length and vocabulary; function-word habits are stubborn. That stubbornness is what makes the method work in both directions.