The fight against ChatGPT’s obsession with the em dash may finally be over. For nearly two years, users of OpenAI’s chatbot have complained that no matter how explicitly they instructed it not to use long, dramatic punctuation, it couldn’t help itself. Writers begged, editors despaired, and the Internet began to regard them as the infallible signature of ChatGPT.

Now, OpenAI CEO Sam Altman says the company has finally tamed the punctuation rebel. Visiting X, he announced a “small-but-happy victory” for users frustrated by the AI’s stylistic stubbornness. “Small-but-happy victory: If you tell ChatGPT not to use em dashes in your custom directives, it finally does what it’s supposed to!”. Altman wrote confirming that the model had finally learned to obey one of the simplest (yet most ignored) user requests.

ChatGPT finally got over the em dash obsession

When tested, ChatGPT appeared to be truly compliant. When asked to write without the em dash, he responded, “Understood! I will avoid using the em dash in all responses.” Furthermore, this setting is also saved in memory to follow it in future responses. For many writers, this is the kind of obedience they have been waiting for since 2022.

This issue may seem minor, but for editors and AI enthusiasts, it became one of the chatbot’s most notorious quirks. Even when users added explicit instructions, “Do not use em dashes”, ChatGPT still included them liberally throughout their text. The punctuation problem turned into a meme, with the em dash becoming shorthand for “Yes, it’s AI-generated”.

By fixing this, OpenAI has done more than refine the grammar, indicating that the chatbot’s “custom instructions” feature, designed to make writing styles more controllable, is finally getting sharpened. Users can now set tone preferences, style notes, and even quirks like punctuation habits and expect to hear exactly what ChatGPT says.

That said, some users responding to Altman’s announcement claimed that the fix has not yet reached their accounts, with some reporting that ChatGPT is still occasionally skipping dashes. Others joked that the punctuation victory was good, but the model was still confusing the facts, such as confidently naming Joe Biden as US president long after the election cycle had changed.

ChatGPT becomes more social with group chat

While punctuation purists celebrated, OpenAI launched something big: group chat for ChatGPT. The new feature adds a social layer to the AI ​​experience, allowing users to create chat groups with friends or colleagues and collaborate with the chatbot.

It works much like WhatsApp or Slack, users can create a group, create a link, and invite up to 20 people to join. ChatGPT then acts as a built-in assistant, capable of brainstorming ideas, answering questions, or mediating conversations. This is a big step forward for OpenAI, transforming ChatGPT from a single experience to a shared workspace.

The feature is still in pilot mode and has only been launched in select regions including Japan, New Zealand, South Korea, and Taiwan, but will be expanded to other countries soon. The company says the rollout is designed to collect initial feedback before a full-scale launch.

Group chat follows a series of experimental launches from OpenAI as it moves beyond text-based chatting. In recent months, the company has introduced Sora, an AI-powered app for creating and sharing short-form videos similar to TikTok or Instagram Reels, and Atlas, its AI browser for Mac users designed to rival Google Chrome and Brave.

Between group conversations, social video apps, and long-awaited punctuation improvements, OpenAI intends to prove that no improvement is too big or too small.