Two pieces of writing can carry the exact same information, yet one reads like it came from a person while the other feels slightly off. Pick apart the awkward one line by line and the grammar checks out, the vocabulary is fine. Still, something feels hollow once you finish reading.

Tone and style make the difference. The smallest choices shape the biggest impressions. This piece maps out the patterns that give writing its "AI smell" and the concrete fixes for each one. It applies to solo publishers, content writers, and anyone who writes for their business.

"AI Smell" Isn't a Matter of Taste

Let's be clear about one thing first. The peculiar awkwardness of AI-written text isn't just a labeling problem — "it feels off because AI wrote it." There are linguistically measurable patterns behind it.

A study by Carnegie Mellon researchers published in PNAS analyzed writing from GPT-4o and Llama 3 and found the following: even when instruction-tuned large language models are asked to write in a casual, conversational register, they use nouns, determiners, and prepositions at markedly higher rates than humans do. They use adjectives, adverbs, personal pronouns, and epistemic markers — hedges like "I think" or "perhaps" — less often.

Put simply, AI leans toward a noun-heavy, information-dense style and fails to reproduce the stylistic variety and personal voice found in human writing. Human authors writing on the same topic form a wide, heterogeneous cluster, while LLM output clusters tightly by model, showing a striking degree of uniformity.

That uniformity is what "AI smell" really is. Follow AI-style patterns even as a human writer, and you'll produce the same awkwardness. Conversely, strip those patterns out of an AI-generated draft, and it reads as human.

Eight Patterns to Eliminate

There are recognizable patterns that give writing its AI smell. Learn them, and you'll spot exactly where to fix your own drafts.

Pattern 1: Grandiose Declarations

This is the most common and most damaging pattern of all. A sentence makes a sweeping claim with no evidence to carry its weight.

Weak: — "AI is changing everything." — "We're living in an age of revolutionary transformation." — "It produces overwhelming results." — "It redefines the very essence of publishing." — "This is the heart of the future."

Why do these fall flat? Because the size of the claim outstrips the specifics backing it up. The sentence reads smoothly but says nothing.

Strong: — "AI is changing things in this specific area." — "A different approach is now possible." — "Thirty Korean publishers have adopted this tool."

Scale down the claim and replace it with a concrete example or number. A modest statement earns more trust than a grand assertion.

Pattern 2: Nominalization

This is the habit of expressing an action as a noun rather than a verb. It's a common byproduct of stiff, translated phrasing.

Weak: — "conduct a review" — "hold a meeting" — "make an effort" — "carry out a decision" — "conduct an analysis" — "the importance of the change"

Strong: — "review" — "meet" — "try" — "decide" — "analyze" — "the change matters"

Render actions as verbs and states as adjectives. The pile of nouns thins out and the sentence gets lighter.

Pattern 3: Stiff, Translated-Sounding Phrasing

This is phrasing that reads as though it were carried over word for word from another language — constructions no one would actually say out loud.

Weak: — "Our company strives to provide our valued customers with the best possible service." — "This study sought to derive findings regarding ~." — "The following items are included herein." — "This is a matter of considerable importance."

Strong: — "We want to give our customers the best service." — "This study tried to find ~." — "Here's what's included." — "This matters."

Swap English-style passive voice and stiff formality for active voice and natural, everyday phrasing.

Pattern 4: Monotonous Sentence Rhythm

When every sentence runs the same length, the writing goes flat. It's the single most telling feature of AI-generated text.

Weak (everything the same length): "The AI era has arrived. Tools are evolving. Users are changing too. Markets are being reshaped."

Strong (varied rhythm): "The AI era is here. As tools evolve and users change alongside them, the whole market's landscape is being reshaped all over again."

Or:

— "Tools are changing. Fast. And the ripples from that change are showing up everywhere in the market at once."

Deliberately mix short sentences with long ones. Even a one-word sentence is fair game. Sentence rhythm is what gives writing its breath.

Pattern 5: AI-Style Openings and Closings

AI has a signature way of opening and closing a piece. It's become so familiar that a single line is enough to give it away.

Weak openings: — "Today, we live in an age of ~." — "Recently, ~ has been drawing attention." — "Let's take a look at ~." — "In this piece, we'll examine ~."

Weak closings: — "In conclusion, we must ~." — "We look forward to future change." — "All of this means that ~." — "In closing, ~."

A good opening starts with something concrete — a scene, a quote, a number, an event. "After deciding to launch an AI startup, Kwak Seok-young, a Stanford MBA student, wrestled with whether to stay in Korea or move to the US." A first line like that pulls the reader in.

A good closing compresses the whole piece into a single line without adding new information. "The bottleneck was never the code. It just looked like code for fifty years."

Pattern 6: Overusing Transitional Words

This is the habit of stitching every sentence to the next with connectors like "therefore," "as a result," "moreover," "furthermore," and "meanwhile," leaving no gaps.

Weak: "AI tools are advancing. Therefore, users are growing in number. Moreover, the market is expanding. As a result, new opportunities are being created. Furthermore, this affects the entire industry."

Strong: "AI tools are advancing. Users are growing in number, and the market is expanding. New opportunities are emerging. The whole industry is caught in that ripple."

The logic still holds without the connectors. If anything, the more of them you use, the more mechanical the writing sounds. If a paragraph has two or more, make a conscious effort to cut them.

Pattern 7: Excessive Deference

Even within an appropriately polite register, over-the-top deference sounds forced.

Weak: — "We humbly ask that our esteemed readers please do this." — "We wish to graciously inform our respected readers." — "We humbly recommend this to our valued customers."

Strong: — "Try this." — "Here's some guidance for readers." — "We recommend this to customers."

Stay polite, but keep the phrasing direct and clear. That builds warmth without losing weight.

Pattern 8: Fusing Words Together

This one is specific to Korean spacing conventions: artificially collapsing the space between words in a compound noun to make it look like specialized jargon.

Weak: — "digitaltransformation" — "AItool" — "contentcreator" — "userexperience"

Strong: — "digital transformation" — "AI tool" — "content creator" — "user experience"

The squeezed-together spelling is meant to look like technical terminology, but it reads as awkward to native readers. Natural spacing makes text easier to read.

Five Traits of Good Writing

Removing bad patterns isn't enough on its own. There are traits worth actively pursuing too.

Specificity

Use concrete examples, numbers, quotes, and scenes instead of abstract nouns.

Weak: "Companies are rapidly adopting AI."

Strong: "Roughly 320 Mega MGC Coffee franchise owners have filed a class-action suit against the parent company."

Abstraction doesn't stick in the mind. Specifics do. The same message lands and stays with readers once it's been run through a concrete example.

Everyday Vocabulary

Favor native words over borrowed, Latinate-style formal vocabulary, and plain phrasing over difficult terminology.

Weak: "We must pursue profit maximization through operational efficiency."

Strong: "Tighten up operations bit by bit, and margins come back to life."

Same meaning, different rhythm. Everyday words hit harder.

The Writer's Own Voice

Carry a quoted speaker's words over exactly as said, but fill the connective tissue between quotes with your own voice. The interplay between quotation and the writer's own voice is what gives a piece its depth.

Variation in Rhythm

Mix short sentences with long ones. A firm statement followed by elaboration, elaboration followed by a short, firm statement. This rhythm is what makes writing feel alive.

Machines produce uniform rhythm. People produce rhythm that shifts with their breath. Build rhythm into your writing deliberately, and the natural feel comes back.

The First and Last Sentence

These two spots decide the impression the whole piece leaves.

Strong opening-line patterns

A concrete scene — "Kwak Seok-young, a Stanford MBA student, is..."

A provocative claim — "Code was never the bottleneck."

An eye-catching number — "$40 billion is more than double the Korean government's entire annual R&D budget."

An unexpected fact — "More and more restaurants are opting out of AI."

Strong closing-line patterns

— "The bottleneck was never the code. It just looked like code for fifty years." — "If you don't understand the cost structure behind a 1,500-won cup of coffee, someone else is designing that structure for you." — "Today's market is telling us that the time left to find that path keeps shrinking."

Don't add new information. Compress the entire piece into one sentence. Leave the reader with a single line that stays.

A Checklist for Reviewing Your Own Writing

Here's a practical way to check whether your own writing has that AI smell.

1. Reread your first sentence. If it opens with "Today," "Recently," or "Let's take a look at ~," rewrite it. Start over with a concrete scene or a number instead.

2. Mark your transitional words. Highlight every "therefore," "as a result," "moreover," and "meanwhile" like they're under a highlighter pen. If a paragraph has two or more, cut half of them and reread. If the flow still holds without them, cutting them was the right call.

3. Count your sentence lengths. If every sentence in a paragraph runs about the same length, deliberately rewrite some short and some long. Build in rhythm.

4. Check the size of your claims. Look for words like "revolutionary," "overwhelming," and "fundamental." See whether the same paragraph has a concrete example or number to justify them. If not, cut the word or add the evidence.

5. Turn nouns back into verbs. Find phrases like "carry out a ~," "conduct a ~," or abstract "-ness/-ity" nouns, and convert them into verb or adjective forms.

6. Read it out loud. Check whether it flows naturally when spoken. Wherever you stumble is where it needs work.

7. Revisit your first 500 characters and last 200. These decide the impression the whole piece leaves. A clichéd opening keeps people from reading on even if the body is strong. A clichéd closing scatters an otherwise good message.

Just running through these seven checks deliberately changes the surface of your writing.

Writing as a Tool

One last point needs making. Writing is no longer just self-expression — it's become a tool for business and work.

A column by a solo entrepreneur. A book description by a publisher. A report by a planner. Every one of these pieces is a channel for carrying your work out into the world. If the writing smells like AI, the work behind it looks lightweight too. If your own voice is alive in the writing, the work comes across with real weight.

The faster AI can churn out text, the sharper the difference human writing makes. Use the tool as a tool, but never skip the final step of polishing it in your own voice. That's the single most important thing about writing in the age of AI.

Reread your own writing. Check how many of the eight patterns above show up in it. Take a pass at fixing each one. Even when the same information sits in the same format, whether your own voice is alive inside it decides the fate of the piece.

Good writing isn't hard. It just has to be deliberate. One conscious pass — noticing what you're writing and how you're writing it — lifts a piece two levels higher.

The faster the tools get, the more the places touched by a human hand stand out. How you leave that mark on your own writing is the central question of writing in the AI era.