Theory: Human writing has natural variation ("burstiness") with some complex sentences and some simple ones. AI tends to produce more uniform, predictable text with consistent complexity throughout.

What we measure: Variance in sentence length and word complexity across the text.

Theory: Type-Token Ratio (TTR) measures unique words vs total words. Humans often have richer, more varied vocabulary. AI may reuse words more frequently.

What we measure: Unique words ÷ Total words, normalized for text length.

Theory: Human writers naturally vary sentence length for rhythm and emphasis. AI often produces sentences of similar length.

What we measure: Standard deviation of sentence lengths.

Theory: AI models often overuse transition words (however, therefore, additionally) to create the appearance of logical flow.

What we measure: Frequency of common transition words relative to text length.

Theory: AI may repeat certain phrases, sentence structures, or filler expressions more than natural writing.

What we measure: Detection of repeated 3-gram phrases and sentence starters.

Theory: AI often uses hedging language (might, could, potentially, generally) to avoid making definitive statements.

What we measure: Frequency of hedging words and phrases.

Theory: Human writing often includes personal opinions and experiences using first-person pronouns. AI tends to be more impersonal.

What we measure: Ratio of first-person pronouns (I, me, my, we, our).

Theory: Humans use varied punctuation including dashes, parentheses, and exclamation marks. AI often sticks to basic punctuation.

What we measure: Diversity of punctuation usage.

Theory: Human writers start sentences with varied words. AI often begins with common patterns like "The", "This", "It".

What we measure: Diversity of first words in sentences.

Theory: Natural language follows Zipf's Law (word frequency distribution). AI-generated text may deviate from this natural pattern.

What we measure: How closely word frequencies match expected Zipf distribution.