How to Tell If AI Wrote Something Before You Send It

Why Your Eye Misses What a Reader Catches

AI-written content isn’t what it used to be with all the “sounds fine to me” thinking flying around. As someone who has drafted close to two hundred client emails using AI assistance, I learned everything there is to know about this particular blind spot. Today, I will share it all with you.

Most of those emails went out fine. Then one didn’t — a prospect wrote back asking if I’d run the pitch through ChatGPT. Felt like getting caught with my hand in the cookie jar, honestly. That’s when it clicked: your own brain is the worst quality control tool you own. You know what you meant. You read your intent into the words rather than the words themselves. It’s like proofreading your own name — you see what should be there, not what actually is.

When you write something from scratch, mistakes feel obvious. A typo jumps out. An unclear sentence nags at you. But when you’re reviewing text an AI generated — text you edited lightly, maybe swapped a word or two — your brain fills in the naturalness that isn’t actually there. The fix is learning to spot AI tells in your own drafts before someone else does. Freelancers lose contracts over this. Managers look less sharp in front of executives. Nobody explains it directly. They just quietly move on.

Sentence Rhythm Is the Fastest Tell

Here’s what AI-generated text actually sounds like when you stop reading for meaning and start reading for sound:

AI version: “Effective project management requires clear communication. Clear communication builds trust. Trust creates accountability. Accountability drives results.”

Every sentence is the same length. Same structure. Same cadence. Rhythmically perfect and utterly robotic. Natural writing — the kind people actually speak and think in — has variation baked in. Some sentences are short. “Like this.” Others ramble because the idea genuinely needs room to breathe. Others have fragments. Fragments that land harder than complete thoughts ever could.

Revised version: “Effective project management starts with clear communication, which builds trust. Trust matters because it creates accountability. And accountability? That’s what actually drives results.”

Different lengths. A question. A fragment. The second version reads like a human thought it — not like an algorithm constructed it at 2am with no coffee. That’s what makes rhythm variation endearing to us writers. So, here’s the test I use: read your draft aloud. Not skimming. Actually read it. If every sentence lands with the same weight and speed, AI probably wrote it. Find yourself naturally speeding up, slowing down, pausing mid-thought? That’s human rhythm. Rewrite the flat sections to match what you hear in your own voice.

Words That Sound Smart but Mean Nothing

AI loves words that sound authoritative while saying almost nothing. I started tracking these after my own drafts kept getting flagged as “too corporate” — even when the ideas were solid. The phrases weren’t wrong exactly. Just filler dressed up as insight.

Here are the actual culprits:

• “In today’s landscape” — Pure throat-clearing. Use “right now” or just cut it entirely. AI defaults here because it’s inoffensive and adds word count without committing to anything specific.
• “Leverage” — Management-speak for “use,” dressed up to sound sophisticated. Say what you actually mean: use, apply, employ. Any of them work better.
• “It is worth noting” — A stalling tactic. Be direct: “This matters because…” or just state the fact and move on.
• “Crucial” — Everything is crucial in AI drafts. Nothing actually is. Reserve it for genuine high stakes, or use “important” and let context do the heavy lifting.
• “Facilitate” — “Enable,” “allow,” “help” are shorter and clearer. AI reaches for “facilitate” to sound more professional. Which is the exact signal that reveals the artifice.
• “Stakeholders” — Fine in formal business writing, but AI overuses it in contexts where “customers,” “team,” or just “people” would sound like an actual human said it.
• “Synergy” — If you’re using this outside a corporate strategy meeting, you’re probably copying from a template. Say what you mean: collaboration, combined effect, working together.
• “Going forward” — Filler. Replace with “next” or cut it. Full stop.

The pattern: AI clusters these at high density. One vague intensifier might slip past. Three in two paragraphs? That’s a tell. Scan your draft for these phrases, highlight every instance, and replace each one with the plainest English word that fits. Your writing gets tighter. Readers stop quietly wondering if you actually believe what you’re saying.

The Structure Test Most People Skip

Okay, here’s the thing nobody tells you. Structure is the meta-tell — while sentence rhythm and word choice catch the reader’s eye, structure reveals whether the thinking behind the words was human or template-based.

AI tends to build things the same way every time: brief intro, exactly three main points of roughly equal length, summary conclusion. Symmetrical. Balanced. Forgettable.

Real writing is messier. Someone explores an idea, realizes mid-thought it connects to something else entirely, doubles back, adds an exception, goes deeper on the part that actually matters. The structure serves the argument — not the other way around. That’s what makes asymmetry feel alive to us readers.

Scan your draft’s outline right now. Count your sections. Are there exactly three? Is each one roughly the same length? Does your conclusion just restate what you already said? That’s AI structure. Break it. Combine two points. Cut a section that’s just there for balance. Add a tangent that genuinely connects to your actual point. Make it uneven. Real thinking is asymmetrical — always has been.

A 90-Second Pre-Send Checklist

Before you hit send, run this routine. Ninety seconds. Catches most of the tells. Don’t make my mistake and skip it because you’re in a hurry.

1. Rhythm check: Read the first two paragraphs aloud. If every sentence lands on the same beat, edit for variation — mix up the lengths, break a long one into two, turn a statement into a question.
2. Phrase scan: Search your draft for “leverage,” “facilitate,” “in today’s landscape,” “crucial,” “worth noting.” Replace each one with plainer language. Every single one.
3. Structure scan: Count your main sections. Similar in length? Symmetrical? If yes and yes, break it deliberately. Combine two points or cut one entirely.
4. Personal detail: Add one specific thing only you would know — a number, a name, a failure, a time you got it completely wrong. AI can’t do this convincingly. I’m apparently the type who learns the hard way, and apparently that detail right there is the kind of thing that signals authenticity instantly.

This habit compounds. The first time you catch a tell before someone else does, the stakes feel real in a way they didn’t before. Your credibility is on you — not luck, not the model you’re using, not the prompt you wrote.