The rule of three, AI, and the quiet crisis in writing craft
Writers love threes. Three reasons, three pillars, three takeaways. It feels tidy, rhythmic, and persuasive. There's even a taxonomy: a triad (three items), a tricolon (three parallel forms), and a hendiatris (three words, one idea).
That pull isn't magic. It's imitation. The French thinker René Girard called it mimetic desire: we copy what admired models do. If Caesar had said four things instead of three, we'd probably swear by a "rule of four."
Why the rule of three works - and when it wears thin
Patterns reduce cognitive load and increase recall. That's good for readers and convenient for writers on a deadline. The problem starts when patterns become templates, and templates become crutches.
One template now dominates AI-assisted drafts: the "X isn't Y-it's Z" move. Linguists call it a contrastive correlative structure with a negated restrictive. It spikes emphasis. It also makes different writers sound the same, fast.
LLMs are mirrors - and megaphones
Large language models train on us. We train on each other. The loop amplifies what's catchy: triads, em dashes, punchy contrasts, familiar analogies. Over time, the market gets flooded with the same tricks.
Call it style hyperinflation. The more a device circulates, the less signal it carries. Readers tune out. Editors reach for the reject button sooner.
What happens when AI writes - and AI reads
Publishing is sliding toward agentic workflows: AI tools that draft, summarize, and trade content with minimal human touch. The full piece is written by AI, skimmed by AI, and "engagement" looks fine on a dashboard.
Researchers warn about "model collapse," where models train on their own synthetic outputs and quality decays into a blur. See the arXiv paper often cited in this debate: The Curse of Recursion: Training on Generated Data Makes Models Forget.
Others argue synthetic data can offset the risk. Maybe. But if everything sounds pre-chewed, writers lose their edge and readers lose trust. The craft suffers long before systems fail.
Practical ways to keep your voice human
You don't need to quit AI. You need guardrails. Here's a short system to keep your work sharp, credible, and yours.
1) Set anti-patterns before you draft
- No default triads. If you use three, prove each item earns its seat.
- Avoid the "not X-Z" contrast unless it's the only clean way to say it.
- Limit em dashes. They're seasoning, not the meal.
2) Replace templates with specifics
- Swap "governments, businesses, civil society" for named actors and actions.
- Use numbers with context: timelines, costs, constraints, trade-offs.
- Show one scene or decision instead of a slogan.
3) Break the rhythm on purpose
- Try two-item or four-item lists to disrupt expectation.
- Vary sentence length: one short line for impact, one longer for nuance.
4) Edit for mimetic tells
- Search for "not," em dashes, and triplets. Cull half.
- Kill clichés and clap-line phrasing. If it reads like an ad, rewrite.
- Remove throat-clearing and vague qualifiers. Say the thing.
5) Use AI as a sparring partner, not a ghostwriter
- Outline in your own notes first. Then ask AI for counterpoints you missed.
- Prompt it to flag patterns you overuse: "Identify contrastive correlatives and suggest plain rewrites."
- Ask for verbs that reduce "is/are." Strong verbs create shape.
6) Build authenticity through the struggle
- Show your constraints: the risk, the trade-off, the doubt you had to settle.
- Make one narrow, testable claim instead of a sweeping generalization.
- Keep a line that cost you effort. Readers can feel the sweat.
Quick diagnostic before you file
- Did you cut at least one triad?
- Did you remove the "not X-Z" structure where possible?
- Is there a concrete example with names, numbers, or a scene?
- Do two paragraphs contain different sentence rhythms?
- Can a skeptic find where you admit a limit or trade-off?
- Did you delete any sentence that sounded like a motivational poster?
A note on the bigger debate
Mimesis is real. That doesn't make originality impossible; it makes it earned. Borrow techniques, but don't outsource the struggle. The thinking you do to write is the benefit you keep, even if the final form uses familiar parts.
If you want a primer on mimetic theory, start here: René Girard. Then get back to the page and test your ideas against resistance.
If you co-write with AI, learn the tools - and the boundaries
Know how models bias phrasing, how prompts steer tone, and how to audit outputs fast. If you're leveling up your workflow, this collection is useful: AI tools for copywriting.
The goal isn't to produce more content. It's to produce work that carries your fingerprint. Choose the harder path at least once in every draft. That choice is the best proof there's a human on the other side of the screen.
Your membership also unlocks:
