ChatGPT Overusing Em Dashes? OpenAI Says a Fix Is Finally Live

Use this Custom Instruction to fix ChatGPT's punctuation tic.

Image: Emiliano Vittoriosi / Unsplash

Have you noticed the excessive use of em dashes by AI models, especially ChatGPT? If you've worked with GPT-4 for any length of time, you've likely seen it—the sudden, unnecessary, and relentless insertion of the em dash. Although this seemed aesthetic and a bit stylistic at first, it later became a recognizable (and often frustrating) AI "tell." Writers, developers, and marketers found themselves in a constant battle, editing out this peculiar AI punctuation tic. Now, OpenAI has finally addressed the "ChatGPT hyphen" epidemic.

It wasn't just that the em dash was present; it was how it was used. The model seemed to favor it as a conversational crutch, often deploying it where a comma, colon, or a new sentence would have been more appropriate. This overuse made AI writing feel less human and more like, well, an LLM default. Due to this excessive use, many writers'—even professional journalists'—work has come under fire for being "AI work," despite not being written by AI. And for professionals, some of you might also fear the stigma of low-effort, copy-paste content, even if you want to integrate AI into a serious workflow.

You don’t need to worry about it. Today, OpenAI officially announced a fix for this em-dash issue, which you are now able to tackle with Custom Instructions, and then ChatGPT will not do it in future responses.

Small-but-happy win:

If you tell ChatGPT not to use em-dashes in your custom instructions, it finally does what it's supposed to do!

— Sam Altman (@sama) November 14, 2025

The CEO of OpenAI, Sam Altman, in the announcement done through his official X handle, explicitly marked this as a “small-but-happy win” for users. The major issue we faced was not only the dash issue but also the inability to remove this excessive dash by prompt, as ChatGPT was previously unable to follow the prompt. The most baffling part of this issue was the model's inability to follow a direct, in-chat instruction to stop. A user might prompt, "Do not use em dashes in your response," only to have ChatGPT reply with something like, "Understood! I will avoid using that punctuation—focusing instead on other ways to structure my sentences." But now this is solved. As the post reads, “If you tell ChatGPT not to use em-dashes in your custom instructions, it finally does what it's supposed to do!”

The Solution: How to Actually Stop the Em Dashes

The ChatGPT update, announced by OpenAI CEO Sam Altman, doesn't remove the em dash from its vocabulary by default. Instead, the fix is implemented through a more persistent tool: Custom Instructions.

On Threads, the official ChatGPT account also shared an image of a "forced apology letter" from GPT for the "ruining" of the em dash, and said, “Chat is now better at not using em dashes—if you instruct it via custom instructions.”

This personalization feature allows you to set preferences and guidelines that ChatGPT remembers across all new chats, effectively re-tuning its default behavior for your specific account.

How to Implement the Fix: A Step-by-Step

To stop the em dash, you need to explicitly tell it to do so in your personalization settings.

• In ChatGPT, click on your name in the bottom-left corner of the screen.
• Select "Personalization" from the menu.
• Choose "Custom Instructions."
• In the box labeled "How would you like ChatGPT to respond?", add a clear, direct instruction.

Example Instruction to Add: "Do not use em dashes (—) in your responses. Use other standard punctuation like commas, colons, or periods to structure sentences."

By adding this directive, you are modifying the model's core behavior for your account, rather than just for a single chat session. Early reports confirm this is significantly more effective than any previous method.

The "Why" Behind the Quirk

While OpenAI hasn't given a full technical post-mortem, the overuse likely stems from the data it was trained on.The em dash is common in certain types of formal and web writing—sources that likely make up a large portion of its training corpus. The model learned to associate this punctuation with "helpful," "nuanced," or "comprehensive" sounding answers and began to over-index on it.

Although the use of em dashes is punctually correct and widely used by numerous writers, either professional or otherwise, its resemblance to AI writing is just a false stigma. However, the overuse of the dash by AI is a correct stigma.