How to Prompt ChatGPT to Stop Using Em-Dashes (And Write Like a Human Instead)

If you read a lot of AI-generated content, you will start noticing a pattern. It is not just the vocabulary or the overly polite tone. It is the punctuation. Specifically, it is the em-dash.

An em-dash is the long dash (—) used to set off a clause, create a dramatic pause, or replace parentheses. Humans use them occasionally to add emphasis or change direction in a sentence. ChatGPT, however, uses them constantly.

If you ask ChatGPT to write a blog post, a product description, or an email, you will likely find sentences structured like this:

"Our software simplifies your accounting process-saving you hours of tedious manual data entry-so you can focus on growing your business."

While grammatically correct, this pattern becomes exhausting to read when it appears in every single paragraph. It makes your drafts look choppy, breathless, and obviously written by an artificial intelligence.

You can stop this habit at the source. This guide will show you how to prompt ChatGPT to write with clean, natural punctuation, explain why the model relies on em-dashes in the first place, and give you practical tools to clean up your existing text.

The Quick Fix Prompt

If you want to stop ChatGPT from using em-dashes immediately, you do not need to read a long guide on linguistics. You just need to give the model a clear constraint before it starts writing.

Copy and paste the following style instruction into your prompt, or add it to your Custom Instructions in ChatGPT:

Do not use em-dashes (—) in the output. Instead, use standard sentence structures. If you need to connect ideas, use coordinating conjunctions (and, but, so), relative clauses, or split the thoughts into separate, distinct sentences. Keep sentences clean and direct.

If you are using a system prompt or building a custom GPT, you can use a more comprehensive instruction set to keep the writing natural:

Punctuation and Syntax Rules:
1. Avoid em-dashes (—) entirely.
2. Avoid overusing semicolons or colons to connect independent clauses.
3. Keep sentences simple and direct. Vary sentence lengths naturally.
4. If a sentence contains more than one parenthetical thought, split it into two separate sentences.

Putting this rule at the beginning of your prompt or in your system settings will significantly reduce the robotic pacing of your AI drafts.

Why ChatGPT is Obsessed with the Em-Dash

To fix this problem permanently, it helps to understand why Large Language Models (LLMs) like ChatGPT write this way. It is not an accident, and it is not because the AI has a personal preference for dramatic pauses. It is a result of how the model processes language.

Linear Token Prediction

LLMs generate text word-by-word, or more accurately, token-by-token. A token is a chunk of characters or a word fragment. When ChatGPT writes a sentence, it does not plan the ending before it writes the beginning. It calculates the most likely next token based on the words it has already written.

This linear generation causes a problem when the model tries to write a complex sentence. If the model starts writing a thought and realizes mid-sentence that it needs to add context or qualify a statement, it cannot go back and rewrite the beginning of the sentence to make it grammatically elegant.

The em-dash acts as a linguistic escape hatch. It allows the model to pause, drop in a secondary thought, and then resume the original sentence structure without having to rewrite anything. It is an easy way to append information on the fly.

The Training Data and RLHF Bias

ChatGPT was trained on billions of pages of internet text. Web writing, especially modern marketing copy, blog posts, and newsletter formats, uses a high volume of em-dashes to create a breezy, conversational tone.

Additionally, ChatGPT was tuned using Reinforcement Learning from Human Feedback (RLHF). During this process, human evaluators rated different AI outputs. Humans tend to prefer text that is easy to skim and sounds enthusiastic. Because em-dashes break up long blocks of text and add dynamic pacing, outputs containing them were rated highly. The model learned that using em-dashes is an easy way to get "good" grades from human editors.

The result is an overcorrection. What works well as an occasional tool for emphasis becomes a tedious habit when applied to every paragraph.

Why the Em-Dash Ruins Readability

When a human writer uses an em-dash, they use it to create a specific rhythm or draw attention to a surprising point. When ChatGPT uses it, it is usually just trying to avoid writing two sentences.

This causes three main problems for your readers.

It tires the reader's eyes

Web readers skim. They look for natural stopping points, which are usually periods. An em-dash forces the eye to pause mid-sentence, jump over a clause, and then pick up the thread again. When every paragraph has two or three of these hurdles, reading becomes tiring.

It sounds overly dramatic

The em-dash is a high-volume punctuation mark. It signals to the reader: Pay attention, this is important. If every minor detail is framed with em-dashes, nothing feels important. It sounds like a late-night infomercial host who never takes a breath.

It is a massive AI footprint

Because almost every default ChatGPT draft relies on this punctuation style, readers, clients, and editors have learned to recognize it. If your blog posts or emails are filled with mid-sentence dashes, your audience will quickly realize they are reading raw AI output.

Before-and-After Rewrites

To see the difference clearly, let us look at some common ways ChatGPT uses em-dashes and how a human editor would rewrite them.

Example 1: The Marketing Pitch

AI Draft: "Our platform offers a centralized dashboard-allowing you to track your team's tasks in real-time-so you never miss a deadline again."
The Problem: The middle clause is forced into the center of the sentence, making the reader hold their breath to get to the end.
Human Rewrite: "Our platform offers a centralized dashboard that tracks your team's tasks in real-time. This ensures you never miss a deadline."
Why it is better: Splitting the thought into two sentences makes the copy feel confident and easy to digest.

Example 2: The Educational Explanation

AI Draft: "Photosynthesis is the process by which green plants use sunlight to synthesize nutrients from carbon dioxide and water-a fundamental chemical reaction that supports almost all life on Earth."
The Problem: The final clause is tacked on as an afterthought, making the sentence drag.
Human Rewrite: "Photosynthesis is the process green plants use to convert sunlight, carbon dioxide, and water into nutrients. This simple reaction supports almost all life on Earth."
Why it is better: The rewrite removes the dramatic dash and uses a clear, active second sentence to state the impact.

Example 3: The Professional Update

AI Draft: "The project is on track for a Q3 release-despite the unexpected supply chain disruptions-thanks to the hard work of our logistics team."
The Problem: The disruption is wedged awkwardly in the middle of a celebratory update.
Human Rewrite: "The project is on track for a Q3 release. Despite unexpected supply chain disruptions, our logistics team kept us on schedule."
Why it is better: By starting the second sentence with the obstacle, the rewrite highlights the team's achievement much more effectively.

Three Advanced Prompting Strategies

If a simple negative constraint does not work, or if you are using an older model like GPT-3.5, you may need a more structured prompting strategy. Here are three ways to train the AI to write with human-grade sentence structures.

Strategy 1: Few-Shot Prompting (Example-Based)

The most effective way to teach an LLM how to write is to show it, not just tell it. This is called few-shot prompting. You provide examples of bad, AI-style writing and contrast them with good, human-style writing.

Try pasting this template into ChatGPT before you ask it to write your content:

Analyze the following examples of writing styles.

Poor Style (AI-style with too many em-dashes):
"Creating content can be difficult-especially when you are short on time-but our tool makes it easy. Just enter your topic-no matter how niche-and get a draft in seconds."

Good Style (Human-style with clean punctuation):
"Creating content is difficult when you are short on time. Our tool simplifies the process. Just enter your topic to get a high-quality draft in seconds."

I want you to write a [insert your topic] using the "Good Style." Do not use em-dashes to connect clauses. Use short, punchy sentences and standard transitions.

Strategy 2: Positive Sentence Structure Rules

Instead of telling the AI what not to do, tell it what it should do. This is often more effective because LLMs sometimes struggle with negative instructions (like "do not use").

Write a guide about [topic]. When structuring your sentences, follow these rules:
- Every sentence should contain only one primary idea.
- If you find yourself adding parenthetical context, start a new sentence instead.
- Use coordinating conjunctions like "and," "but," "or," and "so" to connect simple ideas.
- Aim for an average sentence length of 12 to 15 words.

Strategy 3: The "Print Newsletter" Persona

ChatGPT's default style is conversational web copy. If you change the context of the writing task to a medium that traditionally uses formal, print-style punctuation, the AI will naturally drop the em-dashes.

Write a summary of [topic]. Adopt the tone of an experienced journalist writing for a print newspaper like the New York Times or the Wall Street Journal. Use a formal, objective voice. Avoid conversational punctuation, casual tangents, and modern web writing clichés.

What to Watch Out For: The Punctuation Pivot

When you tell ChatGPT to stop using em-dashes, you will often run into a new problem: the AI will simply swap them for a different punctuation mark.

Instead of writing:

"Our software is easy to use-even for beginners-and saves you time."

It will write:

"Our software is easy to use (even for beginners) and saves you time."
"Our software is easy to use; it is designed for beginners and saves you time."
"Our software is easy to use: it works for beginners and saves you time."

The model is still trying to use the same underlying sentence structure. It is still appending thoughts on the fly, but now it is using parentheses, semicolons, or colons to do it.

To prevent this punctuation pivot, you need to address the sentence structure itself. Your prompts should explicitly limit all of these crutches.

Here is a system prompt that closes these loopholes:

Style Guide: Clean Prose
- No em-dashes (—).
- No semicolons (;) to link clauses.
- Limit colons (:) to introducing lists.
- Avoid parentheses () for inline commentary.
- Keep the syntax simple. If a sentence feels crowded, split it into two sentences.

How to Clean Up Existing Drafts

If you already have a library of AI-generated articles or emails that are filled with em-dashes, you do not need to rewrite them from scratch. You can use a simple editing workflow to clean them up.

1. Use Find and Replace to Spot Them

Do not rely on your eyes to find every em-dash in a 2,000-word draft.

In Google Docs or Microsoft Word, press Ctrl + F (or Cmd + F on a Mac).
Copy and paste an em-dash (—) into the search bar.
Take note of how many times it appears. If you see more than two or three per page, you have an AI-style draft.

2. Apply the "Split and Simplify" Rule

For every em-dash you find, apply one of these three fixes:

The Period Fix: Replace the dash with a period and capitalize the next word. This is usually the easiest and most effective fix.
The Conjunction Fix: Replace the dash with a comma followed by "and," "but," or "so."
The Deletion Fix: Often, the text inside the dashes is filler that can be deleted entirely without losing the meaning of the sentence.

Here is a quick example of the deletion fix:

AI Draft: "Our customer support team is available 24/7-ready to help with any issue-to ensure your business runs smoothly."
Edit: "Our customer support team is available 24/7 to ensure your business runs smoothly." (The phrase "ready to help with any issue" is redundant and can be cut).

Next Steps

The next time you open ChatGPT, do not let it write in its default style. Take ten seconds to set the ground rules.

Add a quick punctuation constraint to your custom instructions, or keep a copy of the "Style Guard" prompt handy in a notepad file. By forcing the model to write without the em-dash crutch, you will instantly improve the quality of your drafts, save time on editing, and produce content that sounds like it was written by a human.