Many readers have come to assume that frequent em dashes are a telltale sign of AI-generated text.
- Altman says ChatGPT will now avoid the em dash if it’s explicitly told to do so.
- With the proliferation of AI-generated and AI-edited text, many readers have come to assume that frequent em dashes are a telltale sign of AI involvement.
- In response to online em-dash memes, some professional writers have dialed back their use of the em dash in their original writing to avoid suspicion.
The em dash was long seen as a sign of mature writing in English, but has lately become an irritant. Its heavy use in AI-generated responses has led several readers to believe that any piece of content with em dashes is produced or edited by AI, even if it’s not, to the point that some writers are avoiding it altogether.
Well, the makers of ChatGPT addressed those concerns. OpenAI CEO Sam Altman said that the company has fixed a glitch, and ChatGPT will now completely avoid em dashes in its responses if explicitly instructed to do so.
“If you tell ChatGPT not to use em-dashes in your custom instructions, it finally does what it’s supposed to do!” Altman said in an X post, calling it a “small-but-happy win.”
It all started with jokes on X and Reddit that frequent em dashes in a piece of text is a quick way to spot bot text, with some calling it “ChatGPT hyphen.” The idea spread fast, fueling a debate.
While the logic that more em dashes indicate potential AI writing is true in some cases, many professional writers have complained of being accused of using AI simply because they like em dashes.
With em dash memes online, readers were quick to label posts, essays, and articles as AI-generated, prompting some writers to intentionally dial them back to avoid suspicion.
One thing to note here is that ChatGPT’s output with em dashes typically omits spaces, whereas some style guides (such as AP) prefer spaces around dashes. The “space vs no-space” became another superficial signal people latched onto.
All in all, the em dash isn’t a very effective way to determine AI-generated text. On the question of why ChatGPT uses a lot of it, some think that perhaps a large amount of ChatGPT’s training data had an em dash. The style tool was frequently used in older books.
For updates and corrections, email newsroom[at]stocktwits[dot]com.<