OpenAI is rolling out a new ChatGPT update that allows users to tell the chatbot to stop using the em dash, a punctuation mark the system has long been known to overuse.
The company said the improvement reflects stronger personalization tools introduced with its latest model, GPT-5.1, although the change has also raised questions about why such behavior persists within large language models.
The announcement came after OpenAI CEO Sam Altman wrote on X that users could now instruct ChatGPT to avoid the punctuation mark entirely.
Altman described the moment as a “small but happy win,” a phrase that quickly spread across social media as users tested the new capability.
In earlier versions of the model, ChatGPT often inserted em dashes in places that many editors considered excessive or stylistically inconsistent.
Attempts to correct the behavior through prompts were not always effective, leaving users frustrated when the model continued using the punctuation.
OpenAI said the ChatGPT update does not represent a global fix in how the model interprets punctuation. Instead, the company has increased the weight of custom instructions, making it easier for users to override default writing patterns.
The shift follows the launch of GPT-5.1, which OpenAI described as more reliable in following preferences and maintaining consistency over long conversations.
Despite that, the company’s announcement also acknowledged that some users still experienced mixed results. In replies to Altman’s post, several individuals shared examples where the chatbot continued to use em dashes even after they attempted to block them.
OpenAI’s explanation highlighted a broader reality while personalization tools can adjust behavior for individual users, the underlying reasons for the model’s stylistic tendencies remain complex and difficult to correct at scale.
Experts said the development reflects the challenges of fine tuning linguistic habits within large language models.
Dr. Mara Ellington, a linguistics professor at New York University, said the update shows the limits of direct model corrections.
“These systems learn patterns from enormous datasets, so something like punctuation habits becomes deeply embedded,” she said. The ChatGPT update essentially teaches the model to listen more carefully rather than altering how it writes by default.
Others said the improvement underscores how companies are leaning heavily on personalization features as a solution to long-standing model quirks.
“OpenAI is addressing stylistic behavior at the user level because adjusting it globally is extremely difficult,” said Victor Hale, a researcher at the Digital Language Institute.
This is a practical workaround, even if it doesn’t solve the underlying cause. Some analysts noted that the development is symbolic of a larger trend in AI deployment.
We are seeing a shift from big AGI promises toward usability enhancements, said Sheila Crawford, an AI governance expert. “These updates may look small, but they reveal where the industry is investing its time.”
Studies have shown that AI writing models consistently display stylistic signatures. A 2024 report from the Center for Computational Writing found that leading language models used em dashes at nearly triple the frequency of professional newsroom editors.
The report also noted a pattern of long, compound sentences in model generated articles, which made the style appear formal but occasionally overly complex.
These tendencies have been linked to the training data, which includes a wide range of writing styles some of which heavily rely on the punctuation mark.
OpenAI has not released data on how often ChatGPT used the em dash before the update, but internal developer notes shared earlier this year suggested that the punctuation habit was a known issue.
The company described it as a “systemic stylistic bias” within its training distribution. By strengthening custom instructions in the ChatGPT update, OpenAI aims to reduce friction for users who want more control over the model’s voice, especially those in journalism, academia or professional communication fields.
Writers, editors and everyday users have shared mixed reactions as they test the update.
“It sounds small, but it makes my workflow smoother,” said Emily Rhodes, an editor at a digital news outlet in Chicago. “We follow AP Style, so unnecessary punctuation can slow down editing. It’s helpful that ChatGPT now follows instructions more precisely.”
However, not all users reported smooth results. “I told it twice to avoid em dashes and it still added one in my first paragraph,” said Akshar Patel, a freelance reporter based in Dubai. “It’s better than before, but not perfect.”
Tech consultant Jordan McBride said the update reveals the complexity behind even minor adjustments. “If fixing punctuation takes this much engineering, it shows how much of a black box these systems still are,” he said.
“People underestimate how difficult it is to overwrite natural language patterns the model learned.” On social media, some users reacted humorously.
“We were promised AGI, and instead we got a punctuation fix,” one X user joked. But others said the improvement aligns with what they expect from modern AI tools.
“Not everything is about AGI,” said another user. “Sometimes people just want cleaner writing.” Analysts expect personalization features to become a major part of AI development in the coming years.
As more professionals rely on AI tools for writing, coding and analysis, companies are likely to focus on giving users more granular control over how the models behave.
Researchers also believe the update could pave the way for more adaptive stylistic control. Future versions may allow users to enforce entire style guides or automatically tune the model to the tone of a publication, newsroom or workplace.
Still, the limitations highlighted by the ChatGPT update suggest that some deeper behaviors remain difficult to adjust without retraining the entire model. OpenAI has not said whether it plans to address the punctuation issue at the model level rather than through personalization.
OpenAI’s latest ChatGPT update offers a targeted improvement that allows users to restrict the system’s em dash use, marking a subtle but meaningful step toward more responsive and customizable AI tools.
The change demonstrates progress in instruction following, even as it exposes the challenges of modifying deeply embedded writing patterns within large language models.
For now, the update stands as an example of how user driven controls continue to shape the evolution of AI systems, reflecting both their potential and their limitations.