“Small-but-happy-win”: OpenAI Finally Fixes ChatGPT’s Most Annoying Quirk, Now No One Can Recognize AI-Generated Text

OpenAI has addressed one of ChatGPT’s most infamous quirks, its excessive use of em-dashes. With the release of GPT-5.1, users can now set custom instructions telling the AI to avoid elongated punctuation, making AI-generated text look more natural. 

Ceo Sam Altman called it a “small-but-happy-win,” highlighting that the update finally allows ChatGPT to follow formatting requests that users have been asking for years.  

“If you tell ChatGPT not to use em-dashes in your custom instructions, it finally does what it’s supposed to do!” he wrote on X. 

Why AI Models Use Em Dashes 

AI models like ChatGPT generate text by predicting the most likely next word or punctuation based on patterns learned from vast amounts of written data. Em dashes (—) are common in formal and creative writing to set off parenthetical information, and to indicate sudden shifts in thought or tone. Since AI models are trained on a mix of human writing that frequently uses em dashes, they tend to overuse them, making the text look “mechanical” or “AI-written.”
Human don’t use em dashes as mechanically, so reducing their overuse makes AI-generated content feel more authentic. 

Social Media Reactions

One user commented, “The fact that it’s been 3 years since ChatGPT first launched, and you’ve only just now managed to make it obey this simple requirement, says a lot about how little control you have over it, and your understanding of its inner workings. Not a good sign for the future.”

Another user commented, “Imagine calling it a “win”  

When the machine finally listens after months of ignoring your soul.”

The third user said, “Great! Hopefully this means that those of us who like em-dashes (the most versatile punctuation mark) eventually can start using them again without being falsely accused of relying on unedited AI writing.”