ChatGPT Em Dash: Why AI Overuses It — and What It Actually Proves
The em dash became the internet's favorite AI tell — people spot two of them in a LinkedIn post and declare it ChatGPT. There is a real pattern underneath the meme, and also a lot of nonsense on top of it. Here is where the habit comes from, what an em dash does and does not prove, and how to check a text properly.
Why ChatGPT loves the em dash
Large language models learn style from their training data, and em dashes are a staple of exactly the prose they were fed most — edited journalism, essays, and web copy, where the dash adds rhythm and a sense of confident flow. Tuning models toward fluent, agreeable answers reinforces it further: the em dash lets a sentence hold two thoughts together without committing to a period. The result is a punctuation frequency few individual humans sustain — noticeable enough that readers turned it into a meme, and real enough that it genuinely is part of the statistical fingerprint of AI writing.
What an em dash does not prove
Humans invented the em dash and plenty of us overuse it — writers, editors, and anyone who grew up on the same prose ChatGPT trained on. Accusing someone of using AI because of punctuation is how people end up falsely accused of using AI. A single tell is folklore. What carries signal is density and company — em dashes in nearly every paragraph, plus the rest of the cluster:
- Tidy three-item lists, everywhere, regardless of whether the topic has three parts
- "It's not just X — it's Y" constructions
- Inflated framing words: delve, tapestry, testament, crucial, pivotal
- Uniform sentence rhythm — every sentence a similar length and shape
- Balanced hedging that never quite takes a side
The full list lives in our guide to AI phrases and tells and how to tell if text is AI-written.
Stop counting dashes — run the real check.
A detector weighs hundreds of statistical signals at once and shows you sentence-level results. Free, no sign-up for your first scan.
Free scan · sentence-level highlighting · no sign-up for your first check
If it's your text: fixing the em-dash habit
- Prompt against it — "no em dashes; use commas, periods, or parentheses" noticeably reduces the rate, though long outputs drift back.
- Edit the rhythm, not just the dash — swapping every em dash for a comma leaves the same uniform sentence shapes. Break some sentences in two; merge others.
- Humanize and verify — our AI humanizer varies punctuation and rhythm in one pass, and the detector then shows you what still scores as AI.
Frequently asked questions
Why does ChatGPT use so many em dashes?
ChatGPT learned to write from huge amounts of edited prose — journalism, essays, marketing copy — where em dashes are a polished-style staple, and its training toward fluent, confident-sounding answers reinforces the habit. The result is em dashes at a rate few individual human writers sustain, which is why readers started noticing.
Does an em dash mean text is AI-written?
No. Em dashes are centuries-old punctuation used by countless human writers — many professional style guides encourage them. One em dash proves nothing. What raises suspicion is density and company: em dashes in nearly every paragraph, alongside other AI habits like tidy three-item lists and "it's not just X, it's Y" constructions.
How do I stop ChatGPT from using em dashes?
Ask directly in your prompt: "Do not use em dashes; use commas, periods, or parentheses instead." It helps but rarely eliminates the habit in longer outputs, so a find-and-replace pass (or a humanizer that varies punctuation) is the reliable fix.
Do AI detectors flag em dashes?
Not as a single signal. Serious detectors score statistical patterns across the whole text — word-choice predictability, sentence-rhythm uniformity, phrase habits — of which punctuation is only a small part. That is why detector results beat punctuation folklore: they weigh hundreds of signals instead of one.
Can I check if a text was written by ChatGPT?
Yes — paste it into an AI detector and read the sentence-level results. Our free checker scored 95.2% accuracy with a 3.1% false-positive rate in a 200-text benchmark and highlights exactly which sentences score as AI, which is far more reliable than counting dashes.
Related: spotting AI writing patterns, what AI slop is, and the best free AI detectors.