Is this a ChatGPT hyphen — or just good writing?

By Dave Lee

As the sophistication of artificial intelligence increases, and we become more disbelieving of our own eyes and ears, it stands to reason that we’ll also grow paranoid that others will think our own hard work is in fact the product of AI.

For example, I was disturbed to learn recently that a growing number of people seem to think the use of the em dash — that short line that allows me to make asides like this one — is evidence of a fraudster. ChatGPT uses the em dash liberally, so it has become a tell.

Things have become so dire that some are inexplicably referring to the em dash as the “ChatGPT hyphen.” Brands that use the mark in their advertising copy get called out as lazy. “Public service announcement,” one popular clip discussing the issue declared, “take out the hyphen.” (Here’s another PSA: It’s not a hyphen.)

On OpenAI’s developer message board, it’s speculated that the bot’s use of the em dash is due to its prevalence in the books and academic papers that OpenAI has ingested into its large language model. That would make sense: The em dash — so named because it is about the width of a capital letter M in whatever typeface is being used — is beloved by writers as one of the most versatile pieces of punctuation at our disposal. It lets you drop in extra information without spoiling the flow. It gives you latitude to write in the way you think, interrupted by thoughts, or other people. Above all else, the em dash is about setting the right pace for whatever it is you are trying to say.

We ran the numbers, and in 173 columns for Bloomberg I have apparently em-dashed 1,357 times. The number would be substantially higher were it not for my editor saving me from myself — a common tussle. In 2011, one New York Times editor gave colleagues a mild scolding after finding no fewer than 16 em dashes on a recent day’s front page. “It can seem like a tic,” he wrote. “Worse yet, it can indicate a profusion of overstuffed and loosely constructed sentences, bulging with parenthetical additions and asides.”

This view — which I clearly don’t subscribe to — means over time the em dash has fallen in and out of favor, as well as in and out of curriculums and grammar guides. But, more to the point, it’s been under technological assault for more than 150 years. The first typewriter patented by Christopher Latham Sholes in 1860s only had room — or so he claimed — for a basic hyphen, forcing typists to either abandon the em dash or opt for its ugly impostor, the double hyphen. As use of typewriters grew, em dash use fell. When digital word processors arrived, the mark remained absent from keyboards, having been considered nonessential. Introducing one meant keying in a cryptic shortcut code, like alt+0151 on Windows. By the time you remembered the right number, any creativity you had for the aside would have long left your body.

Use picked up again thanks to good word processors learning how to automatically turn hyphens or double hyphens into a neatly formatted em dash on a typist’s behalf. After this “ChatGPT hyphen” atrocity, I worry that this automation might fall out of favor as people begin to associate it with AI-generated writing. Already, students are becoming more and more stressed as AI detection tools drum up false positives. One em dash too many and you might find yourself accused of plagiarism.

So now the mark faces yet another evolution at the hands of new technology. Thanks to ChatGPT, my beloved em dash is going from a symbol of sometimes-messy-but-always-human thinking to representing something deeply depressing: the possibility of no thinking at all.