ChatGPT will now comply if you tell it to stop using the em dash (—)

After nearly two years of user complaints, frustration, and countless internet memes, OpenAI has finally solved one of its chatbot's most persistent and infamous stylistic quirks: the seemingly unshakeable obsession with the em dash (—). OpenAI CEO Sam Altman announced the fix on X, calling the successful implementation a "small-but-happy win" for users who have long requested more control over the AI's writing style.

The update ensures that ChatGPT will now honor user preferences specified in "Custom Instructions" to completely avoid the punctuation mark.

The AI's unmistakable signature

For professional writers, students, and editors, the em dash had become the telltale sign of AI-generated content. Even when explicitly told "do not use em dashes," the chatbot would often stubbornly sprinkle the punctuation throughout its output, turning the innocent mark into a digital scarlet letter.

This behavior created an awkward problem. Human writers who naturally favored the versatile punctuation mark found themselves being falsely accused of relying on AI if their text contained too many dashes. The issue was so widespread that the em dash was often jokingly referred to as the "ChatGPT hyphen."

The underlying reason for the obsession is believed to stem from the model's training data, which included vast amounts of 19th-century literature and older web content where the em dash was heavily prevalent. The model simply learned to predict it as the correct token in certain contexts, making the behavior incredibly hard to unlearn.

More than just punctuation

While the fix focuses on a single punctuation mark, the implications are much larger. The successful implementation signals a significant step forward in the personalization and instruction-following capabilities of large language models (LLMs).

The ability to enforce stylistic preferences, however small, is key to integrating AI into professional workflows where maintaining a consistent human voice is paramount. It gives users better control over the "mechanistic" elements of AI-generated prose, helping to make the output less predictable and more authentic.

How to disable the em dash

Users must go into their ChatGPT Custom Instructions (found in the settings menu) and add a clear directive, such as: "Strictly avoid using em dashes (—) in all generated content. Replace them with commas, periods, or parentheses."

While many users have celebrated the announcement, the immediate online reaction has been mixed, with a few early adopters reporting that the ghost of the em dash occasionally lingers, proving that taming a large language model's deep-seated habits is an ongoing process.