OpenAI Ends ChatGPT’s Em Dash Quirk, Rolls Out New Group Chat Feature
OpenAI has finally addressed one of ChatGPT’s most persistent and amusing quirks: its tendency to sprinkle em dashes throughout nearly any piece of writing. For almost two years, users across the globe voiced frustration that the chatbot could not resist this punctuation mark, even when given clear instructions to avoid it. Writers called it distracting, editors called it exhausting, and the internet turned it into a running joke.
This week, the company confirmed that the problem has been resolved. OpenAI CEO Sam Altman announced the improvement on X, describing it as a “small-but-happy win” for users who rely on the model to follow precise stylistic directions. “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!” he wrote, confirming that the issue had finally been tamed.
Initial tests appear to back his claim. When asked to write without the punctuation, ChatGPT reportedly responded, “Got it! I’ll avoid using em dashes in all responses.” The behaviour also persists across conversations, thanks to the platform’s improved custom instruction memory. For many content creators, this adjustment removes a long-standing annoyance and signals that OpenAI is taking fine-grained style control more seriously.
Still, the fix has not rolled out for everyone. Some users replying to Altman noted that they are still seeing the occasional long dash slip through. Others joked that while the punctuation tweak is nice, larger issues remain, including instances where the AI confidently supplies incorrect facts. One user teased that the system still claims Joe Biden is president, hinting at the model’s well-known tendency to hallucinate.
While the em dash victory caught the attention of editors and writing enthusiasts, OpenAI also introduced a much bigger feature: Group Chats for ChatGPT. This addition brings a social dimension to the platform, allowing users to create shared chat spaces similar to those on WhatsApp or Slack. Within these groups, ChatGPT can assist everyone involved, offering ideas, answering questions, and supporting collaborative discussions.
Up to twenty people can join a single group through an invite link. The feature is currently available only in select regions, including Japan, New Zealand, South Korea, and Taiwan. OpenAI says this limited rollout will allow it to gather early feedback before expanding globally.
The new tool arrives as part of a broader push by OpenAI to extend ChatGPT beyond individual text conversations. Other recent releases include Sora, an AI powered short-video creation app, and Atlas, a Mac browser designed to compete with established players like Chrome and Brave.
From social features to punctuation fixes, OpenAI continues to refine both the small details and the larger vision behind its flagship AI, aiming to show users that no improvement is too minor or too ambitious.