humanflow

Stylometry

Stylometry is the identification of an author from measurable habits in how they write — sentence lengths, punctuation, function-word frequencies — rather than from what the writing is about.

Last reviewed 15 August 2026 · The HumanFlow team

In plain English

Everyone has verbal tics they cannot hear themselves using: how often you write "however", whether you use semicolons, how long your average sentence runs. Stylometry measures those and compares them.

It predates AI detection by more than a century, and it answers a different question: not "was this machine-generated" but "is this the same author as that".

A worked example

Stylometry ignores subject matter and looks at the small structural words, because those are the habits a writer cannot consciously control.

Writer 1: "...the data, however, does not support this. It suggests, rather,
          that the effect is small; and small effects are easy to overstate."

Writer 2: "...but the data doesn't support that. It suggests the effect is
          small. Small effects are easy to overstate."

Same claim, same evidence, same register. What differs is entirely structural: writer 1 uses mid-sentence interpolation ("however", "rather") and a semicolon; writer 2 uses short declaratives and a contraction.

A stylometric comparison keys on precisely those features — the frequency of "however", the semicolon rate, the contraction rate — and ignores "data", "effect" and "overstate", because content words change with topic while function words persist across everything a person writes.

Why it matters for AI detection

It is the technique behind the most useful thing a student can do when accused: demonstrating that the disputed work matches their own prior writing. That is a stylometric argument, made informally.

It is also why a portfolio of earlier drafts is worth more than a detection score in an appeal. Comparison against your own known writing is a stronger form of evidence than comparison against writing in general.

Its limits are real. Stylometry needs a decent sample to say anything, style shifts with genre and audience, and it cannot distinguish "written by this person" from "written by someone imitating this person".

Commonly confused with

AI detection
AI detection asks whether text resembles machine output in general, with no reference sample. Stylometry asks whether two specific texts share an author, and needs both.
Perplexity
Perplexity compares text to a model's expectations. Stylometry compares text to another text. One needs a language model; the other needs a writing sample.

Read next

Part of the AI detection glossary.