How to spot AI-written text
I read a great deal of text from across the booth, and lately a certain kind of it has a particular smell. Not bad, exactly — smooth, tidy, agreeable — but strangely weightless, like a room staged for a viewing where nobody actually lives. That’s often AI writing, and while there’s no perfect test, there are reliable tells once you know the shape of the thing.
In the daylight layer I’m a cofounder of Wistkey, so I’ll start with the disclaimer: this is about reading well, not policing. The goal isn’t to catch people — it’s to notice when writing has no one behind it.
The tells
- Too-even rhythm. Human writing is lumpy — a long sentence, then a short one. AI often settles into a smooth, uniform cadence that lulls you and says little.
- Confident vagueness. Lots of sentences that sound like a point without making one: “it’s important to consider various factors.” Fluent, frictionless, empty.
- The tidy triple. Everything arrives in neat groups of three, every section balanced, every list symmetrical. Real thought is messier than that.
- No stakes, no specifics. No odd concrete detail, no strong opinion, nothing a real person would actually risk saying. The safe average of everything.
- Hedge-and-summarize endings. A closing paragraph that restates the intro and reassures you, rather than landing somewhere new.
The tell isn't a word or a phrase. It's the feeling that the room is staged and nobody lives there.
Why detector tools don't work
You’ll be tempted by “AI detector” websites. Don’t trust them. They guess from surface patterns and are wrong constantly — flagging careful human writing (especially by non-native speakers) as AI, and passing polished AI as human. People have been penalized over these false positives. A tool that’s confidently wrong is worse than no tool; it just launders a guess into a verdict.
Read for the person instead
The better test isn’t mechanical, it’s human: is there a specific person behind this?
- Does it say something only this writer, with their particular experience, would say?
- Is there a real example, a genuine opinion, a detail too odd to be generic?
- Does it take a risk — commit to a view someone could disagree with?
And if you’re the one writing with AI and don’t want the weightless smell, the fix is the same from the other side: put yourself back in. Your example, your opinion, your stakes. The tells above are all symptoms of one thing — nobody deciding what they actually mean — which is exactly the part AI can’t do for you.
