In a move that addresses a long-standing complaint about its writing style, OpenAI has announced that its AI model, ChatGPT, will now comply with user instructions to avoid using em dashes. This update, revealed by OpenAI CEO Sam Altman on November 14, 2025, marks the end of what many users have called ChatGPT’s most hated writing habit.
The Rise of Em Dashes as an AI Telltale
Em dashes have long been a telltale sign of AI-generated text, particularly in the output of models like ChatGPT. In April 2025, the term ‘ChatGPT Hyphen’ emerged to describe this pattern. The AI’s heavy reliance on em dashes, often used in place of other punctuation, was flagged as a stylistic quirk that made its text easily detectable in professional and academic settings.
Users expressed frustration at this overuse of em dashes, which they felt detracted from the quality of the text and made it obvious that it was AI-generated. This led to calls for a solution that would allow users to instruct the AI not to use em dashes, a feature that has now been implemented.
Sam Altman’s Announcement on the Fix
OpenAI CEO Sam Altman announced the fix for the em dash issue on November 14, 2025. In his statement, Altman acknowledged that the overuse of em dashes was a key identifier of AI-generated content and that the company had developed a solution in response to widespread user feedback.
Altman’s announcement was met with relief by users who had long complained about the AI’s stylistic habits. The CEO’s acknowledgment of the issue and the company’s swift response to user feedback were seen as positive steps towards improving the quality and usability of AI-generated text.
OpenAI’s Official Update on User Instructions
Following Altman’s announcement, OpenAI confirmed on November 15, 2025, that ChatGPT will now listen to user instructions to avoid using em dashes. This means that users can directly instruct the model to avoid this punctuation in its output. The announcement was also covered by other outlets, with Digital Information World reporting that ChatGPT will listen when people tell it to avoid the long dash.
The implementation of this feature is a significant step towards making AI-generated text more customizable and less easily identifiable as such. It also demonstrates OpenAI’s commitment to addressing user feedback and continually improving its AI models.
Ending ChatGPT’s Em Dash Habit
With this update, OpenAI has effectively ended ChatGPT’s most hated writing habit. As reported on November 16, 2025, users can now remove em dashes from ChatGPT’s output. This change is evident in before-and-after text generation examples, which show a noticeable reduction in the use of em dashes when the AI is instructed to avoid them.
The technical aspects of the update involve tweaks to the model to ensure consistent avoidance of em dashes across responses. This demonstrates OpenAI’s technical prowess and its ability to fine-tune its AI models based on user feedback and requirements.
User Reactions and Broader Implications
The update has been met with positive reactions from users and writers who had previously criticized the overuse of em dashes as a hallmark of ChatGPT’s writing. The relief expressed in late November 2025 coverage reflects the widespread desire for more customizable AI writing tools.
The fix also has broader implications for AI detection tools and plagiarism concerns in industries like journalism and education. By making AI-generated text less easily identifiable, it raises questions about the need for more sophisticated detection tools. However, it also empowers users to generate higher-quality, more natural-sounding text.
Future of AI Writing Customization
This update to ChatGPT’s em dash usage could be a sign of things to come in terms of AI writing customization. OpenAI’s pattern of incorporating user feedback suggests that we may see more features that allow users to customize the AI’s writing style in the future.
Such changes would follow earlier AI evolutions, such as adjustments to tone or length in responses. However, they also raise questions about AI transparency. While fixes like this can make detection harder, they also empower users to generate text that better meets their needs and preferences.
More from MorningOverview