The rule of three: why everything comes in threes -- and why AI learned it the way we did
Third article in the series on rhetorical figures in the age of AI. After the correctio and the mechanics of negation in LLMs, the tricolon -- the three-part figure that structures political speeches and ChatGPT outputs alike.
Sir Winston Churchill, on 13 May 1940, rises to speak before the House of Commons. England is entering total war. He delivers a sentence that will travel across the decades: "I have nothing to offer but blood, toil, tears and sweat" -- preserved in its original form in Hansard (vol. 360, col. 1502).
Blood. Toil. Tears. Sweat. Four terms.
Open any anthology today, any history textbook, any biography of Churchill. The quotation we remember is different: "blood, sweat and tears." Three terms. Toil has vanished. Three beats, three rhythmic accents, and the sentence settles more easily, holds better in memory.
No one made the decision to amputate Churchill. Collective memory did it on its own, without asking. It found four too long, and it tidied up.
This small fact contains the entire thesis of this article. The tricolon -- the three-part rhetorical figure -- is first a cognitive optimum, and a literary ornament only as a consequence. Our brains prefer three because they retain four poorly. The orators of antiquity understood it empirically. Cognitive psychologists confirmed it in the laboratory. Marketing researchers quantified it in advertising. And large language models -- trained on billions of human sentences -- have internalized it to the point of reproducing it to excess.
To understand the tricolon is to understand what unites a political speech, an advertising campaign, a bulleted list generated by ChatGPT, and a Hollywood narrative structure. It is also to know when to break the rule -- because a saturated tricolon is as effective as an over-salted dish.
The figure, as the Greeks named it
The word comes from the Greek τρίκωλον (trikōlon) -- literally "three members." Aristotle, in the Rhetoric (book III, chapter 9), theorizes the periodic rhythm of discourse and favors groups of two or three kôla for their balance and memorability. Cicero, two centuries later, refines the classification in the De Oratore (III, 186) and the Orator (221-226). He identifies in particular the tricolon crescens -- three members of increasing length, which give the sentence its rising momentum.
Quintilian, in the Institutio Oratoria (IX, 3, 77), definitively classes the tricolon among the figures of amplificatio: structures that heighten the impact of an idea through modulated repetition.
The archetype remains unsurpassed. Julius Caesar writes to Rome, after his victory over Pharnaces in 47 BC, the famous dispatch reported by Suetonius: veni, vidi, vici. I came, I saw, I conquered. Three verbs, three syllables each, a perfect crescendo. The message would fit in a single flat sentence -- "I conquered Pharnaces" -- but it would have crossed neither the ancient Mediterranean nor two thousand years of schoolbooks.
Heinrich Lausberg, in his Handbook of Literary Rhetoric (Brill, 1998, §§ 934-947), catalogs hundreds of examples across the Western tradition. The pattern is stable, intercontinental, intertemporal.
The cognitive foundation: why the brain prefers three
George Miller publishes in 1956, in Psychological Review, an article whose title has become legendary: The Magical Number Seven, Plus or Minus Two. In it he demonstrates that human working memory -- the memory that keeps information active during processing -- is limited to about seven items, plus or minus two. The phrase struck so hard that it became one of the clichés of pop psychology.
Except that Miller was more cautious than his posterity. He spoke of chunks -- units of information that depend on the person's expertise. An experienced chess player sees a position as a single chunk; a beginner sees twenty-five.
Nelson Cowan takes up the question again in 2001 with The Magical Number 4 in Short-Term Memory (Behavioral and Brain Sciences, 24(1)). He establishes that, stripped of the artifices of rehearsal and grouping, the effective capacity of working memory is closer to four items. His 2010 revision (The Magical Mystery Four, Current Directions in Psychological Science) confirms this figure across dozens of independent experiments.
Four is the hard limit; three therefore offers operational comfort -- the zone where the brain processes effortlessly, memorizes without rehearsal, reproduces without distortion. Three items sit in working memory with a safety margin; four touch the ceiling; five exceed it.
Churchill said four. Our collective memory rebalanced to three. This transmission is no accident -- it is a cognitive operation of compression, carried out continuously by every brain that has heard or repeated the sentence over the past eighty years.
The tricolon in the history of political ideas
The political formulas that have carried across the centuries almost all obey the rule of three.
Thomas Jefferson drafts the American Declaration of Independence in 1776. He borrows from John Locke the formula "life, liberty, and property" -- and replaces it with "life, liberty, and the pursuit of happiness." The concrete noun becomes abstract, but the ternary structure remains. Two and a half centuries later, the formula is taught in every American school.
Abraham Lincoln, on 19 November 1863 at Gettysburg, reformulates democracy: "government of the people, by the people, for the people." The phrase is itself borrowed from the pastor Theodore Parker (1850), but Lincoln engraves it into History.
The French Republic, through the laws of the Third Republic (14 July 1880), makes official "Liberté, Égalité, Fraternité" -- a formula itself built up by accretion over the course of the nineteenth century (Mona Ozouf, L'homme régénéré, 1989). Liberty alone would be insufficient. Liberty and equality would be binary, and therefore in conflict. The third term -- fraternity -- creates the mediator that links the first two without cancelling them. The ternary structure resolves rhetorically what political philosophy struggles to resolve.
A striking proof of the power of the form: the Vichy regime, in 1940, replaces the Republican motto with "Travail, Famille, Patrie" (Work, Family, Fatherland). The content changes radically, the ideology flips, but the ternary structure remains intact. Pétain could have chosen any formulation. He took up the mold. Because no other mold would have held in collective memory with the same efficiency. Regimes change, the tricolon remains.
The same logic appears among the Scouts de France since 1911: "Franchise, Dévouement, Pureté" (Candor, Devotion, Purity). Three cardinal virtues. In the Catholic religion, the sign of the cross structures the founding prayer around a triad: "In nomine Patris, et Filii, et Spiritus Sancti" -- three names that scan a gesture more than fifteen centuries old.
Cinema does not escape the rule either. Sergio Leone, in 1966, titles his culminating western Il buono, il brutto, il cattivo -- The Good, the Bad and the Ugly. Three characters, three archetypes, and Ennio Morricone's theme takes up the structure across three notes, the last of which stretches out to mark the close. One could multiply the examples -- Olympic (Citius, Altius, Fortius), safety (Stop, Look, Listen), educational. Every culture produces its tricolons. None produces them by chance.
The marketing proof: Shu and Carlson, 2014
For a long time, marketing specialists had the intuition that three arguments were worth more than two or four. The intuition was waiting for its empirical proof.
Suzanne Shu and Kurt Carlson provide it in 2014 in the Journal of Marketing (vol. 78, no. 1, pp. 127-139) with an article bearing a programmatic title: "When Three Charms But Four Alarms: Identifying the Optimal Number of Claims in Persuasion Settings."
Their protocol is methodical. They vary the number of arguments in advertising messages -- one, two, three, four, five, six arguments -- then measure perceived persuasion, purchase intent and, the key finding, the level of skepticism triggered in the recipient.
The results are clear. Persuasion rises up to three arguments. Beyond that, it falls. But the fall is not a passive saturation -- it is an active activation of skepticism. The fourth argument triggers in the reader a critical reflex: "why so many arguments? what are you trying to hide from me?" The fifth and sixth accelerate the rejection.
Shu and Carlson name this effect persuasion backfire. Too many arguments persuade less than three -- and persuade against rather than for.
The lesson is precise and counterintuitive. In an executive summary, a sales pitch, a pitch deck, do not offer four reasons. Do not offer five. Three is the optimum. And the research puts a number on the effect: each argument beyond three reduces conviction by roughly twelve percent.
AI and the tricolon: massive overuse
If you have read a hundred outputs from ChatGPT, Claude or Gemini, you have already noticed the structure. AI loves to state things in threes. "Clear, concise, convincing." "Analyze, structure, deliver." "Simple. Effective. Testable." Bulleted lists come in three points. Conclusions rest on three pillars. Automatically generated slogans almost always obey the rule.
Liang et al. (2024, Stanford, arXiv:2403.07183) document this over-representation in their study Monitoring AI-Modified Content at Scale. The overuse of listable and ternary structures is among the stable stylometric markers of AI writing, alongside the correctio.
Where does this tendency come from? Two hypotheses combine.
The first concerns the training corpus. LLMs are fed human texts in which the tricolon is already over-represented, because orators, marketers and writers have used it intensively for millennia. The model reproduces what it has seen.
The second concerns alignment through RLHF (Ouyang et al., Training language models to follow instructions with human feedback, arXiv:2203.02155, 2022). The human annotators who evaluated the outputs of the alignment models prefer, at equal content, ternary formulations. This preference is injected into the model through reinforcement learning. The model learns that a list of three is more pleasing than a list of four -- precisely because it is true for humans.
AI has amplified a cognitive preference we have shared all along, to the point where this preference becomes recognizable as a stylistic artifact.
When the tricolon fails
The rule of three is not universal. It has documented exceptions worth knowing.
Hebrew biblical poetry, as analyzed by James Kugel in The Idea of Biblical Poetry (1981), operates massively on binary parallelism. "The heavens declare the glory of God, / and the firmament proclaims his handiwork" (Psalm 19:1). Two members; Hebrew oral culture favored this binary repetition as a mnemonic device.
De Mooij, in Global Marketing and Advertising (2010, chapter 6), observes that high-context cultures -- Japanese, Chinese, Korean -- often prefer structures of two or five elements, never three. The Western three can sound schematic, simplistic, foreign in those cultures.
Christopher Booker, in The Seven Basic Plots (2004), identifies seven fundamental narrative arcs -- deliberately above the magical number four, to signal that narration exceeds short working memory and mobilizes longer-term schemas.
The tricolon dominates short, argumentative Western rhetoric as a local optimum -- valuable within its perimeter, discreet elsewhere.
Practical implications: dosing it
What holds for the correctio holds here: recognizing the power of the figure does not license imposing it everywhere.
In an executive summary, three benefits, three risks, three steps. Four begins to saturate, five triggers the skepticism documented by Shu and Carlson. An executive summary with eight numbered points betrays the writer who has not arbitrated -- and activates in the evaluator the critical reflex of the fourth argument and beyond.
In a commercial proposal, limit lists to three items. If you have four strong arguments, arbitrate -- either merge two of them, or drop the weakest. The temptation to say everything dilutes the message.
In a spoken pitch, three benefits. Always. What does not fit in three will not hold in the memory of the person across from you after the meeting.
In a LinkedIn post or an article, be careful not to stack tricolons. Three successive tricolons in the same text produce the immediately recognizable "generic AI" effect. One tricolon per major section is enough to set the rhythm without saturating.
When to refuse the tricolon: before an audience from a high-context culture (Asia, traditional Hebrew cultures), before material that calls for the binary (oppositions, paradoxes), or before an honest list that naturally has four or five items. Artificially compressing it to three betrays reality.
What the tricolon reveals
The tricolon works because it espouses the shape of human cognition. It predates writing, it will outlast LLMs, it has structured thought ever since articulate language has existed.
Greek and Latin orators had identified the optimum through use, two millennia before Nelson Cowan measured it in the laboratory, three millennia before Shu and Carlson quantified it in advertising, and three and a half millennia before GPT reproduced it through statistical imitation.
Our machines learned this law by observing us. They now apply it with the relentless rigor and occasional excess proper to their nature. The challenge, for the human who writes, is to keep it as a tool without turning it into a tic. To know when three serves the point, and when the point calls for one or two or seven.
Churchill said four. Memory kept three. This compression is neither error nor betrayal -- it is a cognitive optimum that imposed itself against the orator's will. To know this is to know how to write to be remembered.
The next article in the series will explore anaphora -- the initial repetition that structures the speeches of Martin Luther King as much as the most effective prompts. Another classical figure that, like the tricolon, is found simultaneously in ancient rhetoric, the neuroscience of memory, and the architecture of attention systems.
Principal sources
- Aristotle, Rhetoric, Book III, chapter 9 (theory of periodic rhythm and the kôla).
- Cicero, De Oratore, III, 186; Orator, 221-226 (theory of the tricolon crescens).
- Quintilian, Institutio Oratoria, IX, 3, 77 (classification among the figures of amplificatio).
- Lausberg, H. (1998). Handbook of Literary Rhetoric (trans. of Handbuch der literarischen Rhetorik, 1960). Brill. §§ 934-947.
- Miller, G. A. (1956). The Magical Number Seven, Plus or Minus Two. Psychological Review, 63(2), 81-97.
- Cowan, N. (2001). The Magical Number 4 in Short-Term Memory. Behavioral and Brain Sciences, 24(1), 87-114.
- Cowan, N. (2010). The Magical Mystery Four. Current Directions in Psychological Science, 19(1), 51-57.
- Gobet, F. & Simon, H. A. (1998). Expert Chess Memory. Memory, 6(3).
- Shu, S. B. & Carlson, K. A. (2014). When Three Charms But Four Alarms: Identifying the Optimal Number of Claims in Persuasion Settings. Journal of Marketing, 78(1), 127-139.
- Ouyang et al. (2022). Training Language Models to Follow Instructions with Human Feedback. arXiv:2203.02155.
- Liang et al. (2024). Monitoring AI-Modified Content at Scale. arXiv:2403.07183.
- Kugel, J. (1981). The Idea of Biblical Poetry: Parallelism and Its History. Yale University Press.
- Ozouf, M. (1989). L'homme régénéré : essais sur la Révolution française. Gallimard.
- de Mooij, M. (2010). Global Marketing and Advertising: Understanding Cultural Paradoxes. Sage.
- Booker, C. (2004). The Seven Basic Plots: Why We Tell Stories. Continuum.
- Churchill, W. -- Speech of 13 May 1940, Hansard vol. 360, col. 1502.
- Lincoln, A. -- Gettysburg Address, 19 November 1863.
- Jefferson, T. -- Declaration of Independence, 4 July 1776.