How to spot AI-written text

The repetition that tries to sound deep

One of the most visible AI tells is stacked repetition. Short sentences, one after another, using the same structure. “It is not about writing more. It is not about posting better. It is not about having more ideas.” The same pattern also appears in positive form, with a series of little punches meant to sound memorable.

A person can use that move well from time to time. The problem is that AI uses it as a shortcut for rhythm when it does not have a real observation to make. Instead of explaining, it hammers. Instead of moving the idea forward, it repeats it in costume.

The fix is not to ban repetition. The fix is to make repetition earn its place. If three sentences are saying almost the same thing, join them into one sentence with a verb and a useful distinction. Instead of three separate blows, write that publishing well does not depend only on frequency, quality, or inspiration, but on having a system that still works when you do not feel like showing up.

Human writing does not need to sound like a motivational talk every four lines. It can have music, of course. But the music should serve the idea, not cover for the fact that the idea has not been worked through.

Colons used as a crutch

Another common pattern is the overuse of colons. AI uses them to create cheap suspense, introduce a conclusion, or make a sentence feel more precise than it is. The result is a draft full of tiny stages where every line seems to be announcing a revelation.

You can recognize the mechanism quickly. One sentence pulls back the curtain, the next delivers the supposed payoff. Then another sentence opens a new curtain, and another delivers again. By the third cycle, the reader is no longer following an idea. They are hearing an automatic cadence.

Colons are useful when they show a clear relationship. They can introduce a list, an explanation, or a sharp clarification. But when they appear constantly, there is usually a cleaner solution. Use a period. Join the sentence. Remove the little performance. Most of those constructions do not lose force when they are simplified. They gain it.

A good test is to read the draft out loud. If you feel like you are presenting headlines inside headlines, there is too much staging. Writing should not keep winking at the reader on every step.

Phrases that announce the turn before making it

AI tends to label the structure out loud. It writes things like “here is where it gets interesting,” “the uncomfortable truth is,” or “what nobody tells you is.” These phrases do not add meaning. They only tell the reader what reaction they are supposed to have.

That stands out because a person with something clear to say does not need to hold up a sign before every turn. If something is interesting, the next sentence will make it interesting. If something is uncomfortable, the discomfort will sit in the specific claim, not in the warning label.

The correction is simple. Delete the signpost and start with the idea. If you were going to write, “Here is where it gets interesting: AI does not fail because it writes badly; it fails because it writes too correctly,” keep the useful thought and develop it without the announcement.

This tell appears a lot in marketing copy, X threads, and LinkedIn posts because it promises quick attention. That is exactly why it deserves suspicion. When every paragraph is presented as a turn, nothing really turns.

Empty hype and decorative language

Some expressions give away a draft because they belong to the packaging, not the idea. “This changes everything,” “this is gold,” “it is not what you think,” “this is a game-changer.” The same goes for fireworks language that replaces an explanation with applause.

A person can be enthusiastic. That is not the issue. The difference is whether the excitement comes from something concrete or whether it has been pasted on top of a weak paragraph to inflate it. Empty hype tries to convince the reader that something matters before showing why it matters.

To fix it, trade intensity for precision. If a tool helps, say where it helps. If an idea feels strong, explain what problem it solves or what decision it changes. If you can only defend a claim with adjectives, the claim may not be ready yet.

It is also worth checking oversized words. “Revolutionary,” “ultimate,” “essential,” and their cousins create debt. They force the sentence to support a much larger promise than the writer has earned. Most drafts improve when the volume goes down and the detail goes up.

Why AI writes this way and how to remove the mask

These tells do not appear because AI is stupid. They appear because it has absorbed huge amounts of writing designed to grab attention quickly. It learns common patterns and reproduces them with confidence. If you do not give it judgment, context, and your own material, it fills the gaps with whatever is most likely.

That is why so many AI-assisted drafts sound similar even when they cover different subjects. They have tidy structure, a decisive tone, and a suspicious lack of friction. What is missing is the odd sentence only that person would write. The small example. The mental order of someone who has actually wrestled with the subject.

The best correction starts before the draft exists. Begin with a specific idea, not a generic topic. Asking for a post about productivity is very different from starting with a voice note recorded after a meeting, while the observation is still fresh and the irritation is still specific. Capturing ideas by talking when they appear helps preserve the raw material before it turns into a pretty but empty sentence.

Then comes the edit. Look for repeated structures, remove rhetorical signposts, replace inflated phrases with details, and ask whether anyone could have signed the piece. If the answer is yes, the voice is not there yet. You do not need to make the writing artificially rough. You need to make it more yours.

Frequently asked questions