ChatGPT’s Em-Dash Overuse and OpenAI’s Update
OpenAI has announced that it has figured out a way to reduce ChatGPT’s tendency to rely heavily on em dashes, addressing a formatting habit that has become widely recognized as a sign of AI-generated writing.
The company explained that although em dashes will still appear when appropriate, ChatGPT is now much better at following user instructions to avoid overusing them.
This update means users can more easily control how frequently the punctuation mark shows up in chatbot responses, especially through personalization settings.
Sam Altman highlighted this improvement on X, saying it’s a small but satisfying fix that finally works as intended when users specify that they don’t want em dashes.
In recent months, em dashes have become a common clue that text may have been produced by an AI system, appearing frequently in schoolwork, emails, and social media posts.
At the same time, many writers have pointed out that the em dash has long been a natural part of human-written prose and shouldn’t automatically be seen as a sign of AI involvement.
There’s still no confirmed technical reason for why language models tend to lean on this punctuation. Some believe it could be linked to older literature included in training datasets, while others speculate that platforms like Medium, which converts hyphens into em dashes automatically, may play a role.
Although OpenAI has attempted to reduce this pattern, the company has not offered specific technical details about how the change was implemented.
How Users Can Reduce Em-Dash Usage in ChatGPT
To help users manage this behavior on their own, OpenAI recommends adjusting ChatGPT through personalization features.
In 2023, the company introduced custom instructions, a tool that lets people define stylistic preferences and guide the assistant’s behavior more consistently.
Users can rely on this feature to limit ChatGPT’s em-dash usage by specifying writing preferences.
- Open ChatGPT.
- On the web version, click your name and then go to Settings, followed by Personalisation.
- Select the Custom Instructions text field to begin entering your preferences.
- Add a request such as: “Please do not use em dashes,” or any other stylistic guideline you prefer.
- Save your changes to apply the new behavior.
This customization feature is available to everyone, including free users, though it is not entirely clear whether it has been fully rolled out in certain regions such as the United Kingdom and European Union.
Also Read:
Texas Secures $40B Google Investment to Power the State’s AI Boom
“7 Best AI Tools That Handle the Hard Work for You, and Some Are Free”
