1. Every sentence is the same length

This is the tell I catch first, and it’s the one most writers miss. Read a paragraph of AI text out loud and you’ll hear it: medium sentence, medium sentence, medium sentence, medium sentence. A metronome. Real people don’t write like that. We write a long, winding sentence when we’re thinking something through — and then a short one. For punch.

The fix is embarrassingly simple: break the rhythm on purpose. Take any paragraph and rewrite one sentence as three words. Then stretch another one out. You don’t need to do it everywhere. Vary the length in even a third of your sentences and the whole piece starts breathing differently, because humans are inconsistent and machines, by default, are not.

Self-test: count the words in five consecutive sentences. If they’re all between 14 and 20, you’ve got the metronome. Break it.

2. The throat-clearing opening

“In today’s fast-paced digital world, content is more important than ever.” I have read this sentence, in one form or another, hundreds of times. It has never once helped a reader. It exists to fill the space before the article starts saying something — throat-clearing dressed up as an introduction.

AI opens this way because it has no opinion about where to start, so it starts nowhere. You do have an opinion. Your real opening is almost always the second paragraph — the one that actually says something. Delete the first one and read what’s left. Nine times out of ten, the article got better and shorter at the same time.

Self-test: does your first paragraph work as the opening of an article about any topic? If yes, it’s throat-clearing. Cut it.

3. The same five words on rotation

Crucial. Seamless. Elevate. Delve. Landscape. AI drafts lean on a small set of favourite words until the whole piece tastes like one flavour. Readers notice, even if they can’t name the words — they just feel the sameness, the way you’d notice a song that only uses three chords.

The fix isn’t a thesaurus. Swapping “crucial” for “pivotal” is still the same sentence wearing a different jacket. The fix is to delete the word and see if the sentence survives without it. “It’s crucial to proofread your work” becomes “Proofread your work.” Shorter, braver, and suddenly it sounds like a person said it. Most of these words are load-bearing for nothing.

Self-test: search your draft for “crucial,” “seamless,” “elevate,” “delve,” and “landscape.” More than one hit each? Delete them all and read it again.

4. Every claim is hedged

“It could be argued that good editing matters.” “This may suggest that planning helps.” AI hedges because it has no stake in the claim. It’s designed to avoid being wrong, so it avoids saying anything firmly at all. The result is writing that sounds like a committee wrote it — every sentence apologising for the one before it.

You have a stake. You tried the thing. You watched it work or fail. Say it straight: “Good editing matters.” “Planning helps.” If you’re wrong, a reader will tell you, and you’ll learn something. A hedged sentence teaches nobody anything, including you.

Self-test: search for “could,” “may,” “might,” and “perhaps” in your draft. Every one you can delete without changing the meaning is a sentence getting braver.

5. Both sides, no position

Here’s the structure AI falls into every single time: the pros, the cons, and a conclusion that says “the best approach depends on your needs.” That’s not writing. That’s a shrug in paragraph form.

Real writing takes a position. “I tried both, and the second one wasted my afternoon.” “Most of this advice is useless for beginners.” “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.

Self-test: read your conclusion. If it could have been written about any topic by anyone, replace it with one sentence that states what you actually think.

6. The summary of the summary

AI drafts love to end twice. There’s the ending, and then there’s a summary of the ending: “In conclusion, by following these steps, you can improve your writing and achieve better results.” Your reader already read the article. They don’t need a recap of the recap.

Worse, the double ending trains the reader to stop paying attention early — they learn that the last paragraph is filler, so they skip it, and then they skip the one before it too. End on the last real point instead. If the piece needs a final line, write one yourself: one sentence, with an opinion in it. Readers remember the last line. Don’t let a template write it.

Self-test: delete your final paragraph. If nothing is lost, it was never doing any work.

7. The specifics vacuum

This is the deepest tell, and the hardest one to fake. AI text is full of claims and empty of specifics. No numbers. No names. No moments. “Many writers struggle with editing” — which writers? How many? When? The draft can’t say, because it never lived through any of it.

One specific detail per section fixes this. A number (“I cut 40% of the draft”). A name (“my editor, Priya, circled it”). A moment (“last Tuesday’s call, when the client went quiet”). AI can’t invent your specifics, so specifics are the fastest way to make a draft unmistakably yours. A reader’s brain reads specifics as proof that a person was there. That’s exactly what they are.

Self-test: pick any section and ask what a reader would remember from it tomorrow. If the answer is “the general idea,” add one specific thing. Just one.

The 60-second self-test

Before you publish, run this on your draft:

• Five consecutive sentences: are they all the same length? Break the rhythm.
• First paragraph: could it open any article about any topic? Cut it.
• Search “crucial,” “seamless,” “delve,” “landscape” — delete the hits.
• Search “could,” “may,” “might” — delete the ones that change nothing.
• Conclusion: does it state what you actually think? If not, rewrite it in one sentence.
• Final paragraph: delete it — was anything lost?
• Every section: does it contain one specific number, name, or moment?

If you fixed three or more tells, your draft already sounds noticeably more human. Fix all seven and it stops sounding generated entirely — not because you tricked anyone, but because you put the human parts back in.

The honest truth

Here’s the thing nobody selling a “humanizer” tool will tell you: robotic writing isn’t a detection problem. It’s a writing problem. Your text sounds robotic because it’s missing the things only a person can supply — uneven rhythm, a real opening, an opinion, a specific moment that actually happened to you.

No tool can add those for you. The tools just shuffle synonyms and inject typos, and weirder is not the same as more human. The fix is the unglamorous part: read it aloud, cut the fog, add one true thing per section. Do that and the draft stops sounding like everyone — and starts sounding like you.