Signs of AI writing — and why plain, careful English trips the same flags

Search “signs of AI writing” and most lists converge on the same nine patterns. They're real — detectors and human readers both key on them. The problem is that careful, simple English shares every one of them, especially when it's written by someone choosing words they're sure of in a second language. This guide lists the nine signs plainly, so you can check your own writing and vary it deliberately — not because the pattern is wrong, but because sounding like yourself matters more than sounding “safe.”

1. Sentences that are all the same length

AI-generated text tends to keep every sentence in a similar range — medium length, similar rhythm, sentence after sentence. Detectors call this low “burstiness”: human writing usually mixes short, punchy sentences with longer, more complex ones (GPTZero). But careful second-language writers often do the same thing on purpose — short, complete sentences are safer than a long one that might go wrong.

2. Heavy formal transitions

Moreover, furthermore, additionally, in conclusion — these are correct, and textbooks teach them as the “proper” way to connect ideas. Used in almost every sentence, they read as generated rather than written. Plain connectors (but, so, also) usually carry the same logic with less weight.

3. Hedging and meta phrases

It is important to note that, it should be noted that, it is worth mentioning — phrases that talk about the writing instead of just writing. They're common in academic English taught as a second language, and just as common in AI output that's been told to sound academic.

4. Suspiciously balanced structure

Exactly three points. On the one hand… on the other hand. A neat conclusion that restates the introduction. Real thinking is usually lopsided — one point matters more than the others, or there's no clean counterargument. Perfect balance is a sign something was generated to a template, whether by a model or by a five-paragraph-essay formula taught in school.

5. No contractions, no informality

Do not instead of don't, throughout. Formal register can be correct for the context — but zero contractions, zero informal touches, in every sentence of a personal essay or email reads as stiff. Non-native writers are often taught that contractions are “less correct,” which produces the same flat formality.

6. Generic claims without specific detail

Technology has changed how we communicate is true of almost anything and proves nothing. Specific, checkable detail — a name, a number, a date, something that could only be about this one situation — is the strongest signal of a real person writing about a real experience.

7. Repeated sentence openers

Starting several sentences in a row with the same word or phrase — Also, … Additionally, … Furthermore, … — is mechanical whether a model does it or a writer relies on a short list of safe openers because they're confident they're grammatical.

8. Vocabulary that's a notch too formal

Utilize instead of use. Commence instead of start. Endeavor instead of try. Dictionaries and translation tools often surface the more formal synonym first, so this pattern shows up constantly in second-language writing — and in AI text trained partly on formal, edited prose.

9. Em dash overuse

The em dash — used to add a clarifying or contrasting thought, like this — has become one of the most-cited “tells” of AI writing since 2023, simply because models use it often. One or two per page is normal style. One per sentence reads as generated, whoever wrote it.

A 2-minute self-check

1. Read it aloud. Uneven rhythm is easy to hear and hard to see.
2. Search for moreover, furthermore, additionally — cut most of them.
3. Search for it is important to note, it should be noted — cut them.
4. Check whether every paragraph has exactly the same shape. Break one.
5. Find one sentence you could make more specific — add a real detail.
6. Count your em dashes. More than one or two per page is worth a second look.