AI writing tells

Formulaic openings

Scroll LinkedIn for five minutes and you'll spot them. "I used to think X. Then Y happened." "The moment the deploy failed, I realized…" "It started with a Slack message." Language models trained on a decade of viral posts learned exactly what a hook looks like, and now they reproduce the same few forever.

Here's what makes this tricky: none of these openings are wrong. Humans invented them, and they worked. That's why they went viral in the first place. The tell isn't the hook. It's that AI reaches for the statistically safest one every single time, lands it with perfectly balanced clauses, and never fumbles. People fumble.

Why detectors flag it

Classifiers have read millions of AI posts opening with the same dozen templates, so the pattern lights up fast. One formulaic opening won't sink you. Stack it on uniform rhythm and an essay-style conclusion, though, and you've completed the signature detectors are trained on.

What it looks like

The day our biggest client churned, I realized everything I knew about customer success was wrong.

Read it twice. Perfectly balanced, isn't it? That's the problem. Real memory doesn't come out this tidy.

How to fix it

• Start in the middle. Skip the setup and open with the concrete detail: "The churn email arrived at 4:47 on a Friday."
• Open with the unpopular opinion itself, not the announcement that you have one.
• Use your own words, even if they're plainer. A flat, specific first line beats a polished template every time.