Home / Patterns / Em dash overuse

Em dash overuse: why the pattern reads as AI-written and how to fix it

The long dash (U+2014) used so often that it replaces full stops, commas and colons throughout the post.

The long dash (U+2014) used so often that it replaces full stops, commas and colons throughout the post. The free checker at klype.ai/check flags it as "Heavy em dash use". It is a lighter finding: the checker deducts less for it than for an invented story or an unsourced figure, since it is quick to fix.

What it looks like

  • We tried everything — new copy, new design, new offer — and nothing moved.
  • The problem wasn't the product — it was the onboarding.
  • Consistency — not talent — is what separates the founders who get leads from the ones who don't.

Example text on this page is written to show the pattern. Figures and people in it are illustrative, not reported.

Why readers distrust it

The long dash became a recognised tell of generated text during 2023 and 2024, because assistants use it far more often than most people do when typing on a phone or in the LinkedIn editor, where the character is awkward to produce. Readers now notice a post with several of them and assume a model wrote it, even when the author has always written that way. The dash also hides weak structure: a sentence that needs three of them usually wanted to be two sentences.

Why it shows up in generated drafts

Edited prose in books and long-form journalism, which makes up much of a model's training data, uses the long dash freely. Models reproduce that frequency in a medium where most people type hyphens or nothing. The mismatch is what readers pick up on.

How the free checker spots it

The checker counts long dash characters (U+2014) in the draft and divides by the word count. More than one per hundred words raises the flag, and the finding reports both numbers. Hyphens and the shorter en dash are not counted. Two long dashes in a 300-word post pass; two in a 150-word post do not.

The checker runs in your browser; nothing you paste is sent anywhere.

How to fix it

  1. Replace each long dash with the punctuation it is standing in for: a full stop where the thought ends, a comma where it continues, a colon where the second half explains the first.
  2. Where a dash pair brackets an aside, decide whether the aside matters. If it does, give it its own sentence. If not, cut it.
  3. Read the result aloud. If a sentence now sounds abrupt, that is usually fine on LinkedIn; short sentences read as written by a person.

Before and after

One draft, rewritten
Before

We tried everything — new copy, new design, a new offer — and nothing moved. The problem wasn't the product — it was the onboarding — and we only saw it when we watched the recordings.

After

We tried new copy, a new design and a new offer. Nothing moved. The problem was not the product. It was the onboarding, and we only saw that when we watched the recordings.

The rewrite removes the pattern without adding anything the author could not stand behind.

Test your draft

The checker reports every pattern it finds, not only this one.

Check drafts against your own voice. The free checker tests the writing. Inside Klype, Check also scores a draft against your own measured voice, matches every figure to the facts saved for your business and applies your publishing rules. Start a 7-day trial.

Common questions

I have always used long dashes. Do I have to stop?

No, but know the cost. Readers cannot see your history; they see a post with four long dashes and apply the association. If the dash is part of your voice, keep it to one per post, where it does the most work, and let full stops carry the rest.

Does the checker count hyphens?

No. It counts only the long dash character, U+2014. Hyphens in compound words and the shorter en dash in ranges are ignored. A draft typed with double hyphens is not flagged by this rule, though some editors convert those to the long dash automatically.

Why does the checker use a ratio instead of a count?

Because one long dash in a 400-word post is unremarkable and three in 100 words is a pattern. The rule flags more than one per hundred words, so the threshold scales with the length of the draft.