Definition
Stylometry is the statistical analysis of writing style used to identify authorship patterns based on vocabulary, sentence structure, punctuation habits, and other measurable features of an individual's writing. In AI detection, it helps distinguish AI-generated text from a specific writer's establi
Stylometry has existed as a field of study for over a century, originally developed to resolve authorship disputes in historical literature. Applied to AI detection, it takes on new significance: if an educator has access to a student’s previous writing, stylometric analysis can identify significant deviations in vocabulary range, sentence complexity, syntactic patterns, and other measurable stylistic features that may indicate AI involvement.
Unlike perplexity or burstiness, which measure how a text compares to AI output in general, stylometry measures how a text compares to a specific writer’s established patterns. This makes it a particularly powerful tool in contexts where comparative samples are available.
How It Works
Stylometric analysis involves extracting measurable features from a text – including function word frequency, average sentence length, vocabulary richness, punctuation patterns, and syntactic structure – and comparing these features across documents. Statistical methods then identify how closely a submitted document matches the author’s established stylometric profile.
In practical academic use, this typically means comparing a suspicious submission against previous assignments by the same student. Significant deviations in vocabulary level, sentence structure, or writing style may indicate that the submitted work was not produced by the same writer.
Why It Matters for AI Detection
Stylometry matters because it addresses a gap that statistical detection signals cannot fill: the question of authorship consistency rather than just AI probability. A document can have elevated perplexity and burstiness – suggesting human writing – while still being significantly inconsistent with the student’s demonstrated writing ability across previous work.
Used in combination with statistical AI detection, stylometric analysis provides a more complete picture of a submission’s authenticity. It is also more resistant to evasion techniques that target statistical signals – it is much harder to mimic a specific individual’s established writing style than to simply vary sentence length.
FAQs
Do AI detectors currently use stylometry?
Some advanced detectors incorporate stylometric features alongside statistical signals. However, stylometry requires comparative samples from the same author, which limits its applicability in detection tools that analyze documents in isolation. It is more commonly used as a supplementary technique when baseline writing samples are available.
Can stylometry definitively prove AI use?
No. Stylometric inconsistency indicates that a submission may not have been written by the named author, but it cannot prove AI use specifically. The student may have significantly improved their writing ability, written in an unusual register, or used a human editor. Stylometric analysis is evidence, not proof.
How much previous writing is needed for stylometric analysis to be reliable?
Generally, stylometric analysis requires at least 1,000-2,000 words of confirmed baseline writing from the same author to produce reliable comparisons. With less text, the stylometric profile is too limited to draw meaningful conclusions. This is why stylometry is most applicable in courses where students have submitted multiple assignments over time.
Can stylometry be used proactively, not just reactively?
Yes. Some institutions maintain stylometric baselines from early-semester assignments to compare against later submissions. This proactive approach is most useful in courses with multiple assessed components – establishing a writing profile at the start of semester makes anomaly detection in subsequent submissions more reliable.