The Rule of Three: AI's Favorite Sentence Pattern

Ask any model to describe a product, and watch what it reaches for: three adjectives, separated by commas, ending on something vaguely powerful. The rule of three is so embedded in AI-generated copy that editors who review it daily can flag suspicious text in seconds just by noticing how sentences land.

This guide explains why it happens, how to catch it in your own drafts, and (more practically) how to rewrite it so the prose sounds like an actual person with a point of view.

Why AI keeps writing in threes

The rule of three has deep roots in rhetoric. Writers have used it for centuries because the human brain is relatively good at holding two comparisons and then receiving a closing beat. AI models absorbed this pattern from an enormous training corpus that skews heavily toward polished, professional writing, which itself skews toward the triplet structure.

The result is that when a model generates a sentence about, say, a coffee brand, it almost automatically reaches for something like "rich, bold, and satisfying." It's not plagiarizing. The pattern is baked into what it learned "good writing" looks like.

What makes this a tell isn't the rule of three itself. The tell is how automatically it appears, and how the items are almost always equivalent in weight and similar in register. Real writers sometimes use the structure deliberately. AI uses it by default.

What the pattern actually looks like

The most common form is the adjective triplet: three descriptors strung together, often with the Oxford comma, often ending on the most abstract of the three. You'll also see it as a verb triplet ("helps you plan, organize, and execute") or a noun triplet describing a product's benefits.

There's a subtler version too. Instead of one triplet sentence, AI often writes three complete sentences on the same topic, each doing roughly equivalent work. The paragraph feels padded, but no single sentence is obviously wrong. Check your draft: if three consecutive sentences cover the same idea from slightly different angles, that's the same impulse at a larger scale.

A quick test: read your draft and count how many sentences land with a third item. If it happens more than twice per page, the pattern is probably doing the structural work that actual reasoning should be doing.

Before/after: fixing a real example

Here's a sentence a language model produced when asked to write an intro for a B2B SaaS landing page:

Before:

Our platform helps teams collaborate faster, communicate more clearly, and deliver better results.

That sentence isn't wrong, exactly. But it contributes nothing. "Collaborate," "communicate," and "deliver" are three ways of saying "work well together," dressed up as specifics. The rhythm announces itself. Any editor will recognize it.

After:

Our platform cuts the back-and-forth that derails projects — the average team saves four hours a week in the first month.

The rewrite drops to one concrete claim. It's less smooth, and that's fine. Smooth is what made the original sound like a press release nobody asked for. A real number, even an imperfect one, anchors the sentence in a way that the triplet never can.

The pattern to internalize: when you find a triplet, ask whether you actually believe all three items are distinct. If they're variations on the same idea, collapse them into one. Then add a detail that earns the space.

Where the pattern hides in longer content

Triplets appear in sentences, but they also appear in paragraph-level structure. AI-generated articles frequently cover exactly three examples, three reasons, three objections. The number three shows up in section headings too, often in "X, Y, and Z" summary heads.

If you're editing an article and notice it has exactly three examples in every section, that's worth looking at. Human writers add a fourth example when they have one, skip examples when they don't, and sometimes drop into a single extended case study instead of distributing attention evenly. The uniformity is the tell. No individual instance of the number three is the problem.

For a broader look at patterns that compound this effect, the 18 signs a piece of text was written by AI guide covers where triplets fit alongside other structural habits.

How to edit the pattern out

The most reliable fix is to decide, before rewriting, what you're actually trying to say. The triplet exists because the model wasn't sure, so it gave you three options dressed as a list.

Pick one. Force yourself to make a single claim, then either defend it in the next sentence or move on. If your claim genuinely requires multiple parts, name them differently. Use contrast, concession, or a sequence with actual progression, not parallel items doing the same job.

The words that instantly signal AI-generated text often cluster around triplets. You'll frequently see "seamlessly," "robust," and "innovative" packed into the same sentence as the three adjectives, which compounds the tell.

A few practical editing moves:

• Find every comma-separated list of three in your draft. Cut one item, or cut all three and replace with a specific detail.
• Check paragraph structure: if three consecutive sentences have similar lengths and cover similar ground, collapse them.
• Read your piece out loud. Triplets have a very specific rhythm that becomes obvious at normal reading speed.

The humanizer prompt at /humanizer-prompt will flag triplet density directly, which saves time when you're editing at volume.

When three items is the right choice

Not every group of three is an AI tell. You use three when you actually have three distinct things that don't reduce to one. A recipe with three ingredients is a list of three because it takes three ingredients. A sentence with three steps is three steps because there are three.

The difference is intention versus reflex. AI-generated triplets usually feel like reflex: the sentence would read the same or better with one item, but the model reached for three because that's what it learned satisfies a reader.

Why AI loves the em dash (and how to spot it) covers a similar mechanism: a rhetorical device that's perfectly legitimate when used deliberately, and deeply suspicious when it appears everywhere at once.

If you find yourself defending a triplet, ask: would removing the third item lose anything? If the answer is no, the third item was never load-bearing.

---

FAQ

Is the rule of three always a sign of AI writing?

No. Writers have used groups of three for a long time, and many human-written pieces use the structure on purpose. The tell is frequency and automaticity. When every other sentence in a piece ends on a third item with that particular rising-then-closing rhythm, you're looking at a model doing what it does by default, not a writer making a deliberate choice.

Why does ChatGPT use triplets so often?

All large language models are trained on text that overrepresents formal, edited prose. That kind of writing uses the rule of three heavily because it's taught as a rhetorical device in schools and style guides. The model learns that triplets appear in "good" writing, so it produces them when trying to produce good writing. It's a pattern-matching artifact, not a stylistic judgment.

How many triplets in an article should I worry about?

There's no hard cutoff, but if you're seeing one per section or more, it's worth reviewing. A 1,500-word piece might have two or three places where three items genuinely belong. If it has twelve, the structure is doing the work that argument should be doing.

Does fixing triplets hurt SEO?

No. Search engines rank content on relevance, authority signals, and engagement. None of that improves because you listed three adjectives instead of one. Replacing filler triplets with specific details generally improves dwell time because the content becomes harder to skim past without reading.

My editor AI keeps reintroducing triplets when I ask it to revise. What do I do?

Add an explicit constraint in your revision prompt. Something like: "Do not use any comma-separated list of three items in this revision." You may need to run the output through a find-and-edit pass regardless, because the reflex is deep. Treating the output as a first draft rather than a finished product helps.