Why your writing sounds like AI, and the nine tells that give it away

Half your company is now required to use AI, and most of them are shipping the raw output. You can tell. Not because a detector flagged it, but because the writing carries a specific set of fingerprints your eye has learned to spot. The good news buried in that: fingerprints are removable. Once you can name the nine tells, you can cut them in ten minutes, and the same draft that read like a machine reads like you.

None of these tells is proof on its own. Plenty of careful human writing uses one or two. What flags a piece as generated is the pileup: three or four of them in a single paragraph, arriving together, in the flat and even way a model produces text when it has no particular point to make. Here they are, worst offenders first.

i.The stock vocabulary

“Delve.” “Leverage.” “Robust.” “Seamless.” “Elevate.” “Harness.” “Tapestry.” “Holistic.” Nobody reaches for these words in a meeting, which is exactly why they read as generated on the page. They're the words a model produces when it's smoothing over a gap where a specific idea should be.

The fix: say the plain word. You don't “leverage” a tool, you use it. Nothing “delves”; it looks.

ii.The stock phrases

"In today's fast-paced world." "At the end of the day." "It's important to note." "When it comes to." These are throat-clearing: they carry no information, and your reader either skims past them or, worse, notices them and starts to wonder who actually wrote this.

The fix: delete the phrase. The sentence left standing is the one you meant.

iii.The em-dash tic

One long dash in a paragraph can be earned. Three in two sentences is a signature, and by now it's a well-known one. If your draft is stitched together with dashes where a comma or a period would do, that alone reads as machine-made.

The fix: swap most of them for a comma, a set of parentheses, or a full stop. Keep the one that earns its place.

iv.The "it's not just X" construction

"It's not just a report, it's a roadmap." "This isn't only about revenue." The antithesis is the single most recognizable AI sentence shape of the moment. It sounds profound and commits to nothing, because the real content is always in the second half and the first half is just scaffolding.

The fix: say the second half. If the roadmap is the point, write about the roadmap.

v.The throat-clearing openers

Sentences that begin "Moreover," "Furthermore," "Additionally," "Notably." Real writers rarely open this way. Models do it constantly, because these words fake the connective tissue that actual reasoning provides.

The fix: cut the opener. If the two sentences don't connect without it, the problem is the sentences, not the transition.

vi.The rule-of-three pileup

"Clear, concise, and compelling." "Faster, smarter, and more efficient." One triad reads as rhetoric. A triad in every paragraph reads as a template filling in its own blanks.

The fix: pick the one word that's true and cut the other two.

vii.The treadmill rhythm

This is the most reliable tell, and the hardest to see, because it's invisible word by word and unmistakable in aggregate. Every sentence the same length, the same shape, the same tempo. Human writing varies on purpose. A long sentence sets something up, and then a short one lands it.

The fix: read it aloud. Anywhere you fall into a rhythm, break it with a short sentence.

viii.The hedge stack

"This could potentially help teams somewhat improve outcomes." Models hedge because they're built to avoid being wrong, but a sentence in which nothing is asserted is a sentence nobody needed.

The fix: decide what you actually think, then write that.

ix.The chatbot residue

Double asterisks around key phrases. Headers and bullet fragments in the middle of what should be a plain email. "I hope this helps!" These aren't writing tells so much as evidence: they mean the text was pasted straight out of a chat window without a single read-through.

The fix: read it once before you send it. That's the whole fix.