Lately, I’ve been running into an editorial dilemma while writing with AI. The model produces a sentence that follows the rule of three: three parallel adjectives or three clauses that escalate neatly. And I immediately like it.
Something like: cold, metallic, unforgiving.
Clean triads. Parallel adjectives. Symmetrical escalation.
I know that this is an AI tell, and so does everyone else who pays attention to the hallmarks of AI writing, but I still want to use it.
Often, each word is doing something. If I delete an adjective or phrase so there are only two items, I don’t like the sentence as much.
AI doesn’t write as naturally as I do for now, but it’s getting close. I’m not afraid to admit it, and this shouldn’t make other writers nervous. I have always said that writers are the best people in the world to get amazing outputs from AI simply due to the nature of what language learning models are. I trust myself to keep the third word when it helps and to delete it when it hurts.
More and more often, I find it helps. Like many other AI writing quirks, the AI is only doing it because it has seen it done well many times in human-written texts, just like the use of the em dash. But AI models are tapping into something beautiful, poetic, and ancient in their use of the rule of threes.
While I’m happy to rephrase sentences so they don’t rely on em dashes, these instincts didn’t originate in a neural network. And when it come to the rule of threes, the instinct is damn good.
Why three works
The rule of three is much older than AI. It’s older than modern English. Classical rhetoric is full of it. Aristotle discussed parallel rhetorical structures as early as the 4th century BCE in Rhetoric, and triadic phrasing shows up constantly in speeches, literature, and advertising.
Life, liberty, and the pursuit of happiness.
Veni, vidi, vici.
Government of the people, by the people, for the people.
Three is the smallest number that creates progression.
One is a statement.
Two creates contrast.
Three creates movement.
The third element gives the structure somewhere to land. It allows escalation or reinterpretation. Triads are memorable because they create a miniature arc inside a sentence.
There’s also a cognitive reason they feel satisfying. Research on working memory suggests people comfortably hold only a few chunks of information at once. Nelson Cowan’s work places the number around four, which means a sequence of three items fits neatly inside that window. I’ve seen AI use the rule of four and loved it too.
Parallel structure also increases what psychologists call processing fluency. When syntax repeats so you have three adjectives with the same grammatical shape, the brain doesn’t have to reparse the sentence each time. It recognizes the pattern and focuses on meaning.
Research on processing fluency shows that information that is easier to process is often experienced as more coherent, more truthful, and more aesthetically pleasing. Triads feel good partly because they reduce cognitive friction while still creating movement inside the sentence.
Two words describe. Three words escalate.
When I’m looking at a phrase like “cold, metallic, unforgiving,” each word contributes something slightly different.
The first establishes orientation.
The second adds texture.
The third interprets.
Cold and metallic describe a surface.
Unforgiving introduces judgment.
That’s why deleting the third word often feels like removing the point. Two words sketch the image. The third tilts the meaning.
You can think about it geometrically. Two points define a line. Three define a plane. With two, adjectives you’re describing along a dimension. With three, you create depth.
Lonely, hungry is a mood.
Lonely, hungry, incandescent becomes something more volatile, more bizarre, more poetic. The third word reframes the first two. So by trying to “fix” a triad by deleting one term, you may be flattening the sentence.
Why AI produces so many triads
If the rule of three is such an effective rhetorical structure, it’s not surprising that AI systems produce it constantly.
As I mentioned and as we all know, language models are trained on enormous corpora of human writing. Edited prose in journalism, essays, speeches, and novels contains a lot of triadic phrasing because it’s persuasive and memorable. Models don’t know the rule of three as a rule. They learn statistical patterns. If similar constructions in the training data often continue with a third adjective, the probability distribution favors completing the triad.
In other words, the system is sampling the next word based on what tends to follow in similar contexts.
This is the core mechanism behind statistical language modeling. Models approximate the probability of the next token given the sequence that precedes it and learning patterns from large corpora of human text. When triadic phrasing appears frequently in edited prose such as journalism, speeches, essays, and fiction, the model naturally learns that completing the pattern with a third term is often the most likely continuation.
If “cold, metallic” frequently becomes “cold, metallic, unforgiving” in human writing, that third slot becomes statistically attractive.
The editing dilemma
This is where the real problem appears. When AI generates a triadic phrase that actually works, I’m caught between two instincts.
The first is stylistic caution. If the triad is too clean, it risks triggering the informal “AI detector” readers now carry in their heads.
The second instinct is poetic. If the sentence reads well—if each word adds meaning, if the escalation is real—deleting one of them feels arbitrary. I’m not tightening the prose. I’m removing a layer.
So the question I end up asking isn’t Does this sound like AI? It’s something more precise. Does the third word actually change the sentence? If it’s redundant, it goes. If it reframes the first two, it stays.
Pattern vs. voice
This is also where voice enters the picture.
AI can produce unusual metaphors. Sometimes they’re genuinely striking. Occasionally they feel like clever combinations assembled from distant domains rather than something emerging from the internal logic of the piece.
When I revise, I’m not counting adjectives or banning triads. I’m looking for conceptual continuity.
Would I think about the world this way? Does this metaphor connect to the themes I keep returning to? Does the escalation make sense in context?
Voice isn’t defined by whether you use rhetorical devices. It’s defined by how those devices connect to your habits of thought. Two writers can both use triads and sound nothing alike.
Keeping the third word
The current trend of treating stylistic patterns as “AI tells” is understandable. People are trying to reverse-engineer how these systems sound. But abandoning rhetorical tools that humans have used for centuries would be a strange response.
The rule of three works because it aligns with how people process language. It creates rhythm. It creates escalation. It gives sentences a place to land.
Sometimes two words are enough. Sometimes the unfinished phrase is stronger. But sometimes the third word is exactly what the sentence has been building toward.
When that happens, deleting it just to make the sentence look less machine-like feels like the wrong kind of edit.
If a triad is empty decoration, cut it, sure, but if each word is earning its place, don’t be scared to keep all three. The goal of writing isn’t to avoid patterns. It’s to use them deliberately.
Sure, the rule of three is an AI tell. It’s also rhetoric doing what it has always done. Writers have no reason to fear it.
