12 Tells That Give Away AI Writing

Readers have gotten good at spotting AI text, often without knowing what tipped them off. Detectors look for many of the same things. This lesson names the 12 patterns that come up most, drawn from published research on AI vocabulary and from the guide Wikipedia's editors keep for spotting AI-written articles.

One tell on its own means nothing. Good human writers use every one of these sometimes. What gives text away is when several of them pile up in the same few paragraphs.

The 12 tells at a glance

Word-level tells (1 to 3)

These are the easiest to spot and the easiest to fix. Studies of scientific papers found some words jumped in use right after ChatGPT came out. One analysis of PubMed abstracts found delves rose from almost nothing in 2020 to over 14 uses per million words in 2024. GPTZero reports that the phrase play a significant role in shaping shows up about 182 times more often in AI text than in human text.

Inflated importance is the cousin of stock vocabulary. AI likes to tell you that things matter instead of showing how. A town stands as a testament to its enduring heritage. A product plays a crucial role in modern workflows. Ask what actually happened, then write that.

Transition openers are habits from school essays that models picked up and overdid. Most of the time the sentence works fine without them. Lesson 2.3 gives you a ready-made banned-word list to paste into your prompts.

Sentence-level tells (4 to 9)

These are harder to see because each sentence looks fine. The pattern only shows across a paragraph.

"Not just X" sentences try to sound insightful by correcting a claim nobody made. Readers now recognize the shape instantly, especially on LinkedIn.

Lists of three feel complete and balanced, which is why models love them. Real writers usually have one main reason, maybe two. When every list in a piece has exactly three items, it reads like a template.

Even rhythm is the burstiness problem from lesson 1.2. Em dashes aren't wrong, and plenty of good writers love them. But AI uses them so often that a page full of them now draws suspicion. Use them when you mean them.

Hedging comes from training that rewards balanced, inoffensive answers. Trailing "-ing" clauses are the same instinct: a sentence ends, then adds a vague comment on why it mattered. Both make writing sound careful and empty at the same time.

Whole-piece tells (10 to 12)

No specifics is the biggest one, and the hardest to fake. AI can't know your numbers, your customers or what happened last Tuesday. Text that never mentions a name, a date or a figure reads as generic because it is. Lesson 3.3 is all about fixing it.

The summary ending repeats what the reader just read. People rarely do this outside school essays. Stop when you've made your last point.

Chatbot leftovers are lines the AI wrote to you, not to your reader. They're the most embarrassing tell and the most certain. The AI detector marks them in red and treats any of them as a strong sign of pasted AI text.