You don't have to guess anymore.

Text A's four sentences run 10, 9, 26, and 8 words, a real spread, some short, one long. Text B's three sentences run 11, 25, and 26 words, and the last two land within one word of each other. Human writing wanders. AI writing tends to settle into a length and stay there.

Count your own sentence lengths for one paragraph. A few short, a couple long, no pattern, that's a normal human spread. If most of your sentences land within a word or two of each other, that's worth a second look.

AI leans on stock connective phrases, "in other words," "that said," "here's the thing," to bridge ideas. A human usually just cuts to the next sentence. Neither Text A nor Text B below actually uses one of these, so this tell isn't demonstrated on this exact pair, it's a real tell, just not provable on these two texts. Compare: "Mornings are rough, especially Mondays. In other words, the whole week starts on the back foot" reads like a stock bridge. "Mornings are rough, especially Mondays. The whole week starts on the back foot" is how a person usually just says it.

Scan any draft for those bridge phrases. One isn't a tell by itself. Three or four in a row usually is.

Human writing tends to carry one small, weirdly specific, personal detail, a name, a smell, a place, a number that's oddly precise. AI writing stays general. Text A's closest attempt is "Monday mornings," real, but too generic to count as a clean example, and Text B stays fully abstract throughout, so this tell isn't cleanly demonstrated on this pair either, it doesn't get forced. Compare: "I dread mornings" is general. "I dread mornings because the dog starts barking at 6:03 sharp" is the kind of detail AI rarely invents.

Look for one concrete, specific, slightly odd detail. If a paragraph is all general statements and zero specifics, that's worth a second look, from a human or an AI.

Detectors score patterns. They don't know who actually wrote something. A wrong guess has real cost, students and employees get flagged off a bad score. The fix isn't a better detector, it's knowing the actual tells yourself.

Use this as a gut check on a detector score you don't trust, not as proof of anything. Pattern literacy is decision support, not a verdict.

1. Count your sentence lengths for one paragraph and look for real variation. 2. Scan for stock transition phrases: in other words, that said, on the other hand. 3. Look for one small, specific, lived in detail. If it's all general, that's worth a rewrite.

Run this on your next email or caption before you send it. Three checks, about two minutes, no tool required.

Em dashes doing a period's job. A tidy summary bow at the end of every paragraph. Resume words nobody says out loud: delve, leverage, robust, seamless.

These three didn't come up in this video. They're the writer side companion's tells, for fixing your own AI drafts, not spotting someone else's. Full breakdown and the rewrite prompt are on the Humanizer Checklist, linked below.