Start by admitting what the draft is missing

Before I touch a single sentence, I ask one question: what would a person who actually did this thing say differently?

An AI writing about email marketing has never sent a cold email that flopped. An AI writing about productivity has never stared at a deadline at 11pm. That gap is the whole problem — and the whole opportunity. The draft knows the topic; you know the experience. Human-sounding writing is just the topic plus the experience.

So I read the draft once and mark every sentence that could have been written by anyone, about anything. Usually that’s most of it. That’s fine. Those sentences are the raw material, not the finished product.

Give it something real to say

The fastest way to make AI content sound human is to add specifics the model couldn’t possibly know.

I don’t mean research — the draft usually has enough facts. I mean the small, concrete details that only come from doing the thing: the number (“it took me four tries”), the name (“my editor, Priya, circled it”), the moment (“last Tuesday’s call, when the client went quiet”). One specific detail per section is enough. It compounds fast, because a reader’s brain reads specifics as proof that a person was there.

Compare these two sentences:

• “It’s important to proofread your work carefully before publishing.”
• “I once published a post with ‘pubic’ in the headline. It was live for forty minutes.”

The first could have been written by anyone. The second could only have been written by someone with a story. You don’t need embarrassing stories — a number, a time, a place all do the same work.

Read it aloud, then fix what sounds wrong

This is the single highest-leverage test I know, and it takes thirty seconds.

AI writing has a rhythm: smooth, even, and slightly sleepy. Every sentence is roughly the same length. Every paragraph lands in the same place. When you read it out loud, that rhythm becomes obvious — you’ll hear yourself going on autopilot, and that’s exactly where your reader would too.

The fix is embarrassingly simple: break the rhythm. Short sentences. Fragments, even. A one-line paragraph for emphasis.

Real people don’t write in metronomic paragraphs. We trail off. We interrupt ourselves. We repeat a word for emphasis — emphasis. A draft that varies its sentence length by even 30% already sounds noticeably more human, because humans are inconsistent and machines, by default, are not.

Delete the phrases that give it away

Every AI draft leans on the same small set of tics. You know them by now, because you’ve seen them in every AI-written article this year:

• “In today’s fast-paced world” — say which world: “In a week with three deadlines…”
• “delve into,” “tapestry,” “landscape,” “realm” — the actual thing you’re describing
• “Furthermore,” “Moreover,” “Additionally” at the start of every paragraph — just start the paragraph
• “It’s important to note that” — if it’s important, just say it
• “In conclusion” — your reader knows the article is ending

None of these phrases are crimes. They’re fog. And fog is the opposite of voice. When I edit, I don’t replace them with synonyms — I delete them. The sentence underneath is almost always stronger without the throat-clearing.

Let an opinion in

Here’s the one AI can’t do on its own: take a position.

AI drafts hedge. They present both sides, list the pros and cons, and conclude that “the best approach depends on your needs.” That’s not writing — that’s a shrug in paragraph form.

A human writer says: “I tried both, and the second one wasted my afternoon.” Or: “Most of this advice is useless for beginners.” Or simply: “Don’t do this.”

You don’t need to be controversial. You need to be specific about what you actually think. One honest opinion per article changes the entire texture of the piece, because opinions are the one thing the average never contains. The average has no preferences. You do.

Cut 10% and see what’s left

AI overwrites. It says in three sentences what needs one, then adds a summary of the summary. Take any finished draft and cut a tenth of it — whole sentences, not words. What’s left is almost always closer to a human voice, because the human voice was hiding under the extra padding.

I do this cut last, after the specifics and opinions are in. Cutting first removes the wrong things; cutting last removes the filler around the good things. If cutting 10% breaks nothing, cut a little more.

What not to do

A quick word on the “humanize AI text” tools that promise to make your content undetectable. I’ve tested several, and here’s my honest verdict: they mostly shuffle synonyms and inject typos. They make the text weirder, not more human — and weirder is not the same as better.

Worse, they teach the wrong lesson. The goal was never to fool a detector. The goal is to write something a real person wants to read. If your content only sounds human after running it through a tool that swaps “important” for “crucial,” the problem isn’t the wording. The problem is that nothing in the piece came from you.

Detectors, for what it’s worth, are unreliable anyway — they flag plenty of genuine human writing and miss plenty of machine text. Write for the reader sitting across from you, not for the scanner.

The honest test

Here’s how I know a piece finally sounds human: I read the finished draft and ask whether I’d say any of it out loud to a colleague over coffee.

Not all of it — nobody talks in publishable paragraphs. But if at least a few sentences sound like something I’d actually say, with my examples and my opinions in them, the piece is done. If every sentence sounds like it came from a brochure, I go back and add one more specific thing, one more opinion, one more short sentence.

Making AI content sound human isn’t a trick. It’s just editing with the human parts put back in: specifics, rhythm, opinions, and the courage to delete the fog. Do that, and the draft stops sounding like everyone — and starts sounding like you.