Why Does ChatGPT Use Em Dashes? How to Change Its Style

ChatGPT uses em dashes frequently because they appeared often in the writing it learned from, and because they are a compact way to add a parenthetical phrase or contrast without ending a sentence. The model did not choose them as a stylistic preference; it learned that em dashes appear in a wide range of contexts where a comma, a colon, or a parenthesis could also work. The result is writing where every third sentence interrupts itself with one. Tools like fenbs can hold the writing rules you agree on, so an assistant reads them before it writes rather than after you have already asked for a rewrite.

This is not proof that writing contains AI output. Em dashes are a normal punctuation mark, widely used in journalism, essays, and published books for decades. Finding one in a piece of writing tells you nothing about who wrote it.

Why the model defaults to em dashes

Large language models learn patterns from text. Em dashes appear in many styles of edited English prose: magazine features, academic papers, business writing, and online publishing. The model learned that an em dash is acceptable in a wide range of sentence structures, which makes it a default choice when the model needs to join two related ideas. It is easy to produce, grammatically flexible, and rarely wrong in a strict sense.

The problem is overuse. A single em dash per paragraph reads as deliberate. Four em dashes in two sentences reads as a pattern. When a model produces text at scale, the pattern becomes visible in a way it would not in a single sentence. Readers who encounter it often describe the writing as robotic or formulaic, even when each individual use is technically correct.

Is em dash use proof of AI writing

No. Em dashes appear throughout published writing by human authors. Joan Didion used them. The Atlantic uses them. Law review articles use them. A piece of writing that contains em dashes could have been written by a person, an AI, or both. Punctuation frequency is not a reliable way to identify AI authorship.

Automated AI detectors that flag em dashes as evidence of AI writing produce false positives at a high rate. A student who writes naturally with em dashes can receive an incorrect AI authorship flag. Treating punctuation as evidence misunderstands how both human writers and language models actually work.

How to reduce em dashes in ChatGPT output

The most reliable method is a direct instruction in the prompt. The instruction needs to be specific enough for the model to apply. "Write naturally" does not change the default style. "Do not use em dashes" does.

• Add this to your prompt: "Do not use em dashes. Use a comma, a period, a colon, or a new sentence instead."
• If the model uses them anyway: "Rewrite this without any em dashes. Replace each one with a comma, a semicolon, a colon, or a separate sentence, choosing whichever fits the meaning."
• For ongoing work: set a custom instruction in ChatGPT's settings that says "Never use em dashes in any response." Custom instructions persist across conversations.
• If you are asking for a specific piece of writing, name the style: "Write this in the style of a plain business memo: short sentences, no em dashes, no parenthetical asides."

Writing style instructions that actually work

Style instructions work best when they are specific and negative: they name what not to do. "Write clearly" does not change behavior because the model already believes it is writing clearly. "Do not use em dashes, bullet points, or the word utilize" changes behavior because each is a specific, testable constraint.

Build a short list of the style choices that matter most to you. A journalist might say: no em dashes, no exclamation marks, no passive voice, active verbs only. A legal writer might say: formal register, no contractions, citations in the format [1], paragraph breaks after every two sentences. A small number of specific rules produces more consistent results than a long description of a desired style.

When a style rule works, keep it. If you find yourself typing the same instruction in every conversation, it belongs in custom instructions or in a project instruction file rather than in the prompt each time.

When an em dash is the right choice

The goal is not to eliminate em dashes. The goal is to choose when to use one. An em dash is the right choice when you want to mark a sharp break or interruption, introduce an explanation with emphasis, or create a pause that is stronger than a comma but does not end the sentence. One em dash in a paragraph, used deliberately, is a good piece of punctuation. It is the unconscious repetition that makes the writing feel like a pattern.

If you are editing ChatGPT output, look at each em dash individually. Some will be correct and worth keeping. Others will be places where a comma or a new sentence would serve the reader better. The edit is not to find and replace all em dashes; it is to decide whether each one earns its place.

How to do this with fenbs

1. Create a free account and add a project for the writing work.
2. Add the writing style rules your team has agreed on as a skill in fenbs. Name it something like "Writing style rules for [project name]". Include the specific instructions: no em dashes, preferred sentence length, register, and any punctuation choices.
3. Record any style decisions that replaced earlier defaults as Decisions in fenbs. If the team decided to stop using em dashes in all client-facing writing, record that as a standing rule with the person who decided it.
4. Connect your AI assistant to fenbs over MCP from Integrations. Ask it to read the project skills before writing anything. A connected assistant reads the style rules from fenbs at the start of each session.
5. When the style rules change, update the skill in fenbs. All future sessions see the current version.

For stopping a different style pattern in Claude, see the how to stop Claude using em dashes guide.