Suspicious that an awkward email or stilted college essay may have been written by artificial intelligence? Some people think there’s a surefire way to tell — the em dash.
Writers, journalists and other grammar nerds — we’re sorry.
In recent months, commenters on social media have zeroed in on the humble punctuation mark — or its overuse — as a sign that writing is produced by generative AI tools like OpenAI’s ChatGPT instead of a human. Bad essays are said to have “a GPT amount of em dashes.”
“LuxeGen,” a fashion podcast aimed at Gen Z listeners, called the em dash a “ChatGPT hyphen” while criticizing a brand’s advertising. “Public service announcement: Take out the hyphen,” a host said.
Not everyone agrees. Most denunciations of the em dash have been met by furious pushback from self-described fans of the punctuation mark, which has traditionally been associated with literary flourishes like Emily Dickinson’s poetry — or the copy of overeager reporters. Novelists and professors said they use the punctuation mark all the time and are loath to cede it to the machines.
“The em dash is such a powerful writing tool that also carries great subtlety to it,” said Aileen Gallagher, a journalism professor at Syracuse University. “The idea that it is an indicator of soulless, dead, AI-generated writing is really upsetting to me.”
Longer than a hyphen, and named after the fact that it is roughly as wide as the letter M, the em dash has long been cherished by writers as a way to pause a sentence or connect ideas with a bit of flair.
“It’s aesthetically elegant,” said J.T. Bushnell, a senior instructor at Oregon State University’s School of Writing, Literature and Film. “It’s a piece of punctuation that’s not out of place in formal context, but it also captures something about natural inflections of speech in a way that other punctuation doesn’t.”
How did it become a polarizing tell for AI users? The search for a silver bullet to identify AI-generated writing has been underway since tools like ChatGPT became available. It’s a tricky science. The chatbots use a complex statistical analysis to predict the words — and punctuation — to string together in a sentence.
Sometimes, there are dead giveaways, like when careless users forget to delete instructions or a chatbot’s dialogue from the text it is asked to generate. But other methods, like AI-detecting tools that claim to analyze text patterns to find evidence of AI handiwork, are imperfect and issue false positives.
So readers are left to make their own deductions about whether ChatGPT writing carries telltale quirks. Some have observed that AI can use unconventional or stilted vocabulary. The word “delve” has been branded an AI buzzword rarely used by humans but favored by chatbots, according to two 2024 investigations of academic papers.
Does the same apply to the em dash?
The accusation began floating around social media early this year. Some would-be AI sleuths claim that the em dash is seldom seen outside of AI text. Others think that chatbots are more prone to misusing or overusing the punctuation mark.
Em dashes are “relatively rare when a human uses it, maybe once or twice, if that,” a post on X said in February. “But AI chats love using it. No clue why.”
We asked the makers of ChatGPT if they agreed. Laurentia Romaniuk, a member of OpenAI’s model behavior team, said it’s possible that ChatGPT-produced writing favors em dashes. But she added that it’s not a hard-and-fast rule — the AI’s output is heavily influenced by how users respond to its outputs and any writing samples users ask it to imitate, factors that are always shifting.
“While we — and ChatGPT — have a soft spot for the em dash, our priority is making sure our models are helping users communicate their ideas clearly and effectively, in whatever style they choose,” Romaniuk said in a statement. “We’re continuing to improve ChatGPT’s writing abilities to this end.”
While the internet deliberates, the em dash’s loyal fans are fuming. Before chatbots entered the picture, writers and editors happily waged their own debates about overusing em dashes. Moniza Hossain, a children’s author based in Britain, called the em dash her “emotional support punctuation mark.”
“We joke about how we use it too much,” she said. “We have to ration it out in our chapters so that we don’t use it too many times.”
“We must protect our sweet little em dash,” said Rebecca Crunden, a fantasy and science fiction writer.
Gallagher, the journalism professor, said magazines and newspapers embraced the em dash in the ’70s as reporters began to write more expressive and narrative stories. Internet bloggers, looking to add voice and verve to their columns, did the same.
That could also explain AI’s fondness for the punctuation mark, she added.
“If (AI) relied a lot on either magazine writing or blog writing, then those two styles were quite fond of the em dash,” Gallagher said.
Authors and artists have protested the use of their work to train AI tools, and Hossain and Crunden said it was galling that a writer’s flourish like the em dash could now be negatively associated with a tool that trained on their words.
Bushnell, the writing instructor, said he hoped the scrutiny AI has brought to the em dash might encourage more people to learn about it and use it correctly in their writing.
Hossain fears it will have the opposite effect. She caught herself refraining from using em dashes in a recent proposal she wrote for a new novel, she said — she didn’t want an editor to suspect her pitch was AI-generated.
“AI robbed the em dash from me,” she joked. “It’s terrible.”