Compulsive tricolons. "I came, I saw, I conquered" works because Caesar used it once. AI uses rule-of-three in every sentence, often stretching to four or five. Colin Gorrie: "What the LLM lacks is not technical ability, but taste — the taste to know when to deploy these techniques."
The Pattern
"Life, liberty, and the pursuit of happiness." "Government of the people, by the people, for the people." Tricolons work. They've worked for centuries.
AI doesn't use tricolons for emphasis. It uses them because three is its default list length. Every list, every time — three adjectives, three bullet points, three clauses. The number never varies because the model never considers whether three items actually serve the point.
Colin Gorrie put it best: AI has the technique of classical rhetoric down cold but zero taste for when to deploy it. A writer saves the tricolon for a moment that earns one. AI treats it as its default sentence structure.
GPTZero's research confirmed what readers already sensed: AI text uses tricolons at rates far above human baselines. Worse, the three items tend to be near-synonyms. "Innovative, groundbreaking, and transformative" is one idea wearing three hats.
It goes deeper than explicit lists. AI prose drifts into triadic rhythms even in narrative — three supporting arguments, three paragraph sections. The number three is baked into the model's sense of what "complete" looks like.
Count the list items. If every list in a piece has exactly three, and the items feel interchangeable, you're reading AI.
Examples
The Research
Gorrie is a classicist and linguist, and his "Why ChatGPT writes like that" (Dead Language Society) remains the sharpest piece on this. He recognized that ChatGPT's rhetorical repertoire is technically fluent but deployed without judgment. The tricolon section is his most cited example, and the quote that stuck: "What the LLM lacks is not technical ability, but taste — the taste to know when to deploy these techniques."
GPTZero ran the numbers. Tricolons in AI text far exceed human baselines, and the items in those triplets tend to be near-synonyms rather than genuinely different ideas. AI doesn't just list three things. It lists the same thing three ways.
tropes.fyi filed it under "Tricolon Abuse." The name says it: the technique is fine, the abuse is the problem. Mechanical repetition kills the device.
VU Amsterdam's ALP Guide ties this to a deeper vocabulary problem: AI text uses 4x fewer unique words than human writing. A small word pool means that when the model reaches for three list items, it keeps grabbing the same ideas in different wrappers.
Caught in the Wild
Three tips. Three lessons. Three takeaways. AI-generated LinkedIn posts default to three-item lists so reliably that some users now deliberately post lists of two, four, or seven as an anti-AI signal.
Originality.ai →"Fast, reliable, and affordable." Marketing copy generators produce this cadence on autopilot. A/B tests show that human-written copy with varied list lengths outperforms the invariant threes — readers pick up on the monotony even when they can't name it.
Writing instructors flag tricolon density as a structural tell. In suspected AI essays, there's a tricolon in nearly every paragraph, and the three items rephrase the same idea in slightly different words. One instructor's shorthand: "If the student suddenly sounds like they're writing a sermon, check for threes."
Sources