How to Make AI Writing Sound More Human

By Frederik Baun, conversion copywriter and founder of Words Sounds Visuals

I've read enough client drafts by now to catch all six of these patterns on sight, and built a free tool that automates the check.

How to Make AI Writing Sound More Human

I already wrote about the words that give AI away. Delve, leverage, tapestry, and the rest of the graveyard are real, and cutting them helps.

But a word list only catches vocabulary. It says nothing about how the sentences around those words are built. That’s where most of the tell lives.

AI builds sentences the same way regardless of topic. Once you’ve seen the pattern, you’ll catch it everywhere, even in drafts that pass every word-list check with a clean sheet.

I’ve read enough client drafts by now to know the six habits on sight, and it’s the same six every time:

• The em dash epidemic
• Binary contrast addiction
• Label-colon openers
• Sentence stacking
• The AI phrase library
• The perfect triple

The first four you can catch inside a single sentence. The last two only show up once you step back and look at the whole page.

The Four Patterns You Can Catch in One Sentence

Read any one sentence in isolation, and these four are all visible on sight.

The Em Dash Epidemic

AI uses the em dash as duct tape. A sentence runs long, and two ideas don’t quite connect. The model welds them together anyway, instead of picking a lane.

Take a sentence like this one: “This platform is built for speed. It also delivers results you can measure.” Somewhere in every AI draft, that period disappears, and a dash welds the two halves together whether they belong together or not.

Delete every dash like that from your draft. A full stop almost always works better, and a comma or parentheses cover the rest.

Binary Contrast Addiction

AI’s other favorite move is to negate a plain claim, then assert a fancier-sounding version of the same idea.

Before: Leadership isn’t about authority, it’s about influence.

After: Influence matters more than authority.

Same idea, minus the theater. The negate-then-assert version feels clever for about half a second. Then you notice it hasn’t said anything the plain version didn’t already say.

Label-Colon Openers

AI also likes to announce the category of its insight before delivering it.

Before: The truth: most founders never escape the trap they built.

After: Most founders never escape the trap they built.

A person doesn’t announce the category of their point before making it. They just make it.

The AI Phrase Library

AI also keeps a small rotation of stock lines it reaches for no matter the topic. It throat-clears with a two-word phrase before any point that supposedly matters.

It also claims secrecy about something that was never secret, and promises urgency with a phrase that stopped meaning anything from overuse.

None of these lines are wrong on their own, but stack two or three in one page, and a reader’s brain flags the pattern before they can say why.

It’s the same instinct that catches the vocabulary graveyard. One word slips past, but the density gives it away.

The Two Patterns You Can Only See Across a Whole Page

These two are different. You won’t catch them in one sentence. You have to sit back and look at the shape of the whole piece.

Sentence Stacking

This is probably the single biggest reason AI writing feels robotic. It happens even when every individual sentence looks fine on its own.

Watch what happens when nothing connects: “Email sequences are a core part of B2B marketing. They require careful planning. Subject lines affect open rates. Timing matters too.”

Every sentence in that paragraph is grammatically fine. And every sentence lands like a tile dropped on a table. There’s no mortar between it and the one before it.

AI doesn’t think in arguments. It predicts the next likely sentence, one at a time. There’s no thread connecting today’s claim to the one it made two sentences ago.

A person writes differently. One thought leads to the next when someone is working through a real point.

So connect what you write: “Email sequences live or die on timing. Subject lines get the open. But if the first email lands on day three instead of day one, it arrives cold. No subject line recovers from that.”

Read both versions out loud, back to back. One sounds like a person thinking. The other sounds like a printer.

And if you’re paying attention, you’ll notice this whole section just connected six sentences about connecting sentences, on purpose.

The Perfect Triple

Three feels complete. It’s one of the oldest tricks in rhetoric, and the human brain never stops falling for it.

AI knows this too, so it applies the triple everywhere, whether three is the right number or not.

Before: Build trust, deliver value, make the ask.

After: Build trust, then make the ask. The value is the trust.

Real thinking doesn’t produce a clean three every time. Sometimes there are two reasons that matter. Sometimes there’s one reason that matters more than the other four combined.

Break a list on purpose. Make it two items, or four. That’s often the tell that a person counted instead of an algorithm deciding for them.

The Six Patterns at a Glance

The same six patterns, condensed into a table worth bookmarking.

The Three-Step Fix I Run on Every Draft

Running a draft through a detector and swapping the flagged words is step zero. It’s not enough by itself.

If you want AI to make writing sound human, run these three checks by hand afterward. Do them in this order.

1. Read the paragraph out loud. Your ear catches disconnected sentences faster than your eyes do, even when you can’t yet explain why nothing connects.

2. Vary your sentence lengths on purpose. Short, then longer, then short again. If three sentences in a row run the same length, fix one of them.

3. Break one list on purpose. When everything comes in perfect threes, cut an item or add a fourth. A person who wrote the list themselves rarely counts that precisely.

Three Real Before-and-After Rewrites

Theory only gets you so far. These three are pulled from the kind of drafts that land in my inbox every week.

Before: “Growth isn’t about more leads. It’s about better ones. Most agencies chase volume. Volume rarely converts.”

After: “Growth depends on lead quality more than lead volume. Most agencies chase volume, and volume rarely converts on its own.”

Before: “The truth: our platform handles everything you need. It covers onboarding. It covers support. It covers billing.”

After: “Our platform covers onboarding, support, and billing, so you’re not stitching together three different tools.”

Before: “The result: clarity, consistency, and conversion.”

After: A coaching client of mine rewrote her homepage using exactly this fix. She told me the page finally sounded like she was in the room, instead of a brochure. Clarity was the only real driver in that rewrite. The other two words in the old version were just decoration.

What This Means for Your Marketing

Go back to the client I mentioned at the start. She’d deleted every flagged word from her landing page, and it still read like a machine wrote it, because nobody had touched the rhythm underneath.

We ran her draft through these six patterns instead of a word list. Four em dashes disappeared, two binary contrasts collapsed into direct statements, and a paragraph of disconnected feature bullets finally got its conjunctions back.

Nothing about her offer changed. The product, the price, and the promise stayed exactly the same, and the only difference was that a person could now read the page and believe someone real had written it.

That belief is worth more than any of us give it credit for. A reader decides whether to trust you within the first few seconds, long before they finish your pitch.

AI rhythm is one of the fastest ways to lose that decision before you’ve even made your case.

I built a free copy cleaner that checks a draft for exactly these six patterns in one pass. Feed it a draft, and it flags the rhythm problems your own eyes have stopped catching.

Rhythm is one layer. If the draft came out of a model, it’s worth checking the claims inside it too.

Fact-checking AI content catches a different failure mode entirely. It flags a confident sentence that also happens to be false. And the full 25-point pass I run on everything before it ships lives in the editing checklist.

That’s the whole difference when you’re trying to make AI writing sound human instead of just passing a checklist. The words were never the hard part. The rhythm is what gives you away, and now you know exactly what to listen for.