OpenAI's ChatGPT Em Dash Fix: A Practical Prompt and Guide to Sound Human

For months, a single piece of punctuation became the scarlet letter of AI writing: the em dash. Its constant, almost compulsive, use by ChatGPT turned into an online joke and the easiest way to call out AI-generated text. While OpenAI has offered a partial "fix," the real issue is the chatbot's robotic tone.

This article gets straight to the point. First, we'll provide a direct, copy-paste solution: a custom prompt to make AI chatbots sound more human. Then, we'll walk you through exactly how to set it up in ChatGPT, Google Gemini, and Claude. After you have the fix, we’ll dive into the background of why this became such a big problem in the first place.

My Go-To Prompt for Ditching the Robot Voice

Tired of fighting ChatGPT's default tone? Me too. After endless rounds of editing out the same cringey phrases, I've settled on a custom instruction that I feed it for almost every task. It’s not magic, but it gets the output about 70% closer to something a human would actually write.

Feel free to copy and paste this directly into your own custom instructions or use it at the start of a new chat.

The Prompt:

Write in a natural, human way. Avoid formulaic structures and the usual AI tells. Do not use the “not X, but Y” pattern. Skip the robotic transitions like “in addition” or “furthermore.” Don’t force lists unless they genuinely make sense. Do not end with cheerful customer-service closers such as “let me know if you have any other questions.” Be concise, direct, and conversational. No marketing tone, no generic optimism, no filler. Produce writing that reads like something a real person would say in a normal conversation, not something generated by a model.

The key is being specific. Instead of just asking it to "sound human," you have to explicitly call out its worst habits. Pointing out the "not X, but Y" pattern and the useless transition words forces it to find other ways to connect ideas. Telling it to skip the cheerful sign-off prevents those annoying, syrupy conclusions.

You’ll still need to edit the output. It’s still an AI, and it will still make weird choices. But starting with a prompt like this makes the difference between a heavy rewrite and a quick polish.

How to Set Up Custom Instructions on Major AI Platforms

To make this practical, here’s how to set your preferred writing style on the most popular AI chatbots so you don't have to type it out every single time.

For OpenAI's ChatGPT (Web & Mobile App):

The process is nearly identical for the website and the app.

1. Log in to ChatGPT.

2. Click your profile name in the bottom-left corner (on web) or open the main menu (on app).

3. Select "Personalization" from the menu.

4. Choose "Custom instructions".

5. You will see two boxes. The first is for information about you. The second box is titled, "How would you like ChatGPT to respond?"

6. Paste the prompt into the second box.

7. Click "Save."

For Google Gemini:

Gemini has a dedicated section for saved instructions, making it easy to set a global preference.

1. Log in to gemini.google.com.

2. In the bottom-left corner, click the Settings gear icon.

3. Select "Instructions for Gemini" from the menu.

4. Click "Add instruction" and paste the prompt into the text box that appears. You can give it a title for your own reference.

5. Click "Save." You can also go directly to this page by visiting: https://gemini.google.com/saved-info

For Anthropic's Claude:

Claude offers a global setting to influence its tone across all your conversations.

1. Log in to claude.ai.

2. Click on your profile initial or picture in the top-right corner and go to Settings.

3. Navigate to the General tab in the settings menu.

4. Find the section titled "What personal preferences should Claude consider in responses?" You can also go directly to this page by visiting: https://claude.ai/settings/general

5. Paste the detailed prompt into this text box and save your changes.

The Deeper Problem: Analysis and Background

Now that you have the tools to fix the AI tone, let's explore how one punctuation mark became such a massive headache.

How One Punctuation Mark Became a Meme

The ChatGPT em dash wasn't just a grammatical tic; it was a cultural tell. Its sudden appearance everywhere led to a weird kind of "punctuation shaming," where writers were accused of being lazy for using a mark that’s been around for centuries. Meanwhile, you couldn't get ChatGPT to stop. This stubborn little habit became the perfect symbol of generative AI's biggest flaw—it’s powerful, but it’s also stubbornly robotic.

OpenAI's "Fix" and Why No One's Convinced

OpenAI’s CEO Sam Altman called it a "small-but-happy win": you can now use custom instructions to tell ChatGPT to drop the em dash. On the surface, great. But the celebration was short-lived. The fix isn't a default setting; it's buried in a feature most casual users will never touch. For a solution, it feels more like a workaround that only a fraction of users will ever implement.

The Em Dash Is Just the Tip of the Iceberg

Let's be honest: the ChatGPT em dash is a symptom, not the disease. The real problem is that AI-generated text still feels fundamentally lifeless and predictable. AI writing has a whole collection of other dead giveaways, from the "Not X, but Y" formula to awkward transitions and annoyingly cheerful sign-offs. These aren't style choices; they're programming habits. They reveal that AI isn't "writing" in the human sense. It's just assembling patterns.

The Unwinnable War on AI "Tells"

This whole ChatGPT em dash episode is the perfect snapshot of our relationship with AI. We're in a constant cat-and-mouse game. As soon as we spot one tell and the developers patch it, we'll just start noticing another one. Human writing isn't about perfectly following rules—it’s about knowing when and how to break them.

So, is the war on AI tells over? Not even close. The real question isn't whether AI can ever perfectly mimic us, but whether we'll trust anything online once it gets close.

Frequently Asked Questions

1. What was the "ChatGPT em dash" problem?

ChatGPT had a bad habit of overusing the em dash (—) in its writing. It became such a common AI trait that people started using it to spot bot-written text immediately.

2. How did OpenAI fix the em dash issue?

OpenAI added an option in the "Custom Instructions" settings. You can now specifically tell the chatbot not to use em dashes, and it should listen.

3. So, is the em dash problem actually solved?

Not really. The fix isn't a default, so you have to know about it and turn it on yourself. Most users won't, so the default AI-generated text will probably still be full of them.

4. What are other dead giveaways of AI-generated text?

Besides the ChatGPT em dash, look for formulaic sentences like "It's not just X, it's Y," robotic transitions, a love for bullet points, and cheesy, overly-helpful closing lines.

5. What does this whole em dash thing tell us about AI?

It shows that AI is still a sophisticated pattern-matcher, not a creative writer. It develops robotic habits that are easy to spot, and fixing one just reveals another.

6. Will AI writing ever be truly undetectable?

Maybe, but we're not there yet. For now, there are still plenty of tells that give it away. The real challenge will be maintaining trust online when AI gets good enough to hide its tracks.

7. Can I go back to using em dashes without being judged?

Go for it. The more real humans use them correctly, the less power the AI stigma will have. It's a punctuation mark, not a crime.