Sam Altman, OpenAI’s CEO, stated on X that ChatGPT will refrain from using em dashes when a user specifies this preference in their Custom Instructions. “It finally does what it’s supposed to do,” he said.
Custom instructions on ChatGPT allow tailoring how the AI responds and can be found in the Personalisation tab in the settings menu.
“Small-but-happy win,” Altman added.
Ever since ChatGPT gained popularity among the masses, the em dash, or ‘—’, has appeared with increasing frequency in online writing, becoming a noticeable hallmark of AI-generated text.
Users have since found it difficult to instruct ChatGPT to avoid using em dashes in a piece of text.
“I pushed, pulled, rewired and begged, but nothing made it stop,” one user wrote in a blog post.
“$100 for anyone who can show me how to get ChatGPT to stop using em dashes. It’s driving me insane,” said Chip Huyen, the author of the popular book ‘AI Engineering’, in a post on X a few months ago.
Multiple factors have been attributed to ChatGPT’s increasing use of em dashes. The model reflects patterns present in its training data, which includes a significant amount of modern prose that relies on the mark for pacing, emphasis and fluid transitions.
“If [AI] relied a lot on either magazine writing or blog writing, then those two styles were quite fond of the em dash,” said Aileen Gallagher, a journalism professor at Syracuse University, in a statement to the Washington Post earlier this year.
Em dashes also offer a compact way to connect or interrupt ideas without additional structure, making them efficient from a token-generation standpoint.
The excess usage of em dashes in ChatGPT has also been a subject of jokes and memes on social media.
Some also say that this correlation is unfair. “It’s annoying that em dashes have become the telltale sign someone used ChatGPT to generate the text,” said one user on X.
“I use them often in my emails and writing—probably incorrectly. Now everyone assumes I’m putting everything through ChatGPT.”
Altman announced this update a day after OpenAI released the GPT 5.1 model. It is an upgrade to the GPT-5 model family that introduces new reasoning features, faster performance on simple tasks, and expanded personalisation tools across ChatGPT.
ALSO READ: EU Data Act Goes Live—Why Today Marks a Turning Point for Enterprise Strategy
