The Small Habits That Give AI Writing Away

You can usually tell before you even think about running something through a checker. There's a feel to it, a slight wrongness that's hard to name at first. After reading enough of it, though, the specific habits start jumping out, and once you see them you can't really unsee them.

Start with the word choices. AI models trained on huge chunks of the internet develop favorites, words that show up constantly in machine text and comparatively rarely in the way actual people talk. Delve. Tapestry. Landscape, used to mean some abstract field rather than an actual place. Boasts, as in "the town boasts a rich history." None of these are wrong exactly. Nobody's going to arrest you for writing "delve." The tell is frequency. A person might use one of these once every few thousand words, if that. A model reaches for the same small set constantly, like it's working off a checklist nobody told you about.

The Compulsive Love of Threes

Then there's the compulsive love of threes. Three examples. Three reasons. Three adjectives in a row before the noun finally shows up. I noticed this pattern in my own writing years before AI models were doing it at scale, actually, which made it strange to watch a machine adopt the exact same crutch I had to train myself out of. It's not wrong to group things in threes sometimes. It becomes obvious when every single list, every single paragraph, follows the same three beat structure without exception.

Rhythm might be the biggest tell of all, and it's the one people notice without being able to explain why something feels off. Human writing swings. You'll get a short sentence. Then something long and looping that takes its time getting to the point, doubling back on itself halfway through. AI generated text tends to smooth that out into something closer to uniform, sentence after sentence landing in the same general length. It reads fine on a technical level. It just doesn't breathe the way writing with an actual person behind it breathes.

"Human writing swings and breathes. AI text flattens sentence length into a uniform metronome."

Formatting Clues and Tone Flatness

Formatting gives it away too, more often than people expect. A heavier reliance on bullet points and bolded subheadings than you'd typically find in something a person just sat down and wrote start to finish. Curly quotation marks turning up where straight ones would normally be, since certain models default to that style regardless of what surrounds it. Small stuff on its own. Adds up fast.

And there's a certain flatness to the tone that's hard to pin down but easy to feel. Competent. Balanced. Weirdly neutral even on topics that would normally pull a strong reaction out of a person. It hedges constantly. It rounds off anything that might read as an actual opinion. You rarely find a joke that lands sideways, or a sentence that could only have come from one specific person's particular way of looking at things.

For what it's worth, Wikipedia's editors have put together detailed internal guidance for catching exactly this kind of writing, because it keeps showing up in articles and it violates several of the site's core content standards. That tells you something. The people who deal with this problem constantly, at scale, still rely on trained human judgment scanning for these patterns rather than trusting any single automated score to make the call.

If you're trying to make your own writing sound like an actual person wrote it, and not like the statistical average of everything the internet has ever said about a topic, this is the list to work against. Vary your sentences on purpose. Use the words you'd actually reach for out loud, not the ones that sound impressive on a page. Let a little mess survive the final edit. That unevenness is usually exactly what makes writing feel like it came from someone real.