I was worried about AI taking my job, maybe even becoming sentient and deciding the human race was surplus to requirements, leading to the Human v AI wars. But no, AI is attacking us humans where it really hurts, our punctuation. (Devious swine)
This devilish attack targets some of the English-speaking world’s favourite punctuation marks, the classic comma, the brilliant colon and semicolon, and (I kid you not) parentheses. But, Mike, what is the em dash? I hear you cry.
According to thesaurus.com, the em dash is seen as being more interruptive or striking than other punctuation, so it is often used stylistically to draw a reader’s attention to a particular bit of information. The em dash resembles a horizontal line (—) that is longer than both a hyphen (-) and an en dash (–).
Here’s an example of using em dashes in a sentence.
They’re everywhere — bloody everywhere — in blog posts, emails, WhatsApps — even bleeding Facebook comments.
You may not have noticed how common the em dash has now become. I noticed it when the use of AI became more common, this observation is backed-up by an actual detective, well a linguistic detective, François Keck. He used an analytical tool to analyse 10,000 English-language scientific journals, he found the relative frequency of em dashes more than doubled between 2021 (before the launch of ChatGPT) and 2025, a magnitude of change unmatched by any other character.
This unprecedented increase in the use of the em dash led to it being given the nickname “the ChatGPT dash”. A not very imaginative or funny nickname, what about the Metal Micky Mark, or the T1000 Tick, or maybe the Data (as in Lt Commander) Dash. (If you can think of any others leave them in the comments.)
The em dash became an AI tell, a giveaway to suggest the writing was AI-generated or at very least formatted by AI. All of a sudden people were writing Facebook comments like:
Before ChatGPT: “omg cant beleve wat govment doin lol ppl wake up ???”
After ChatGPT: “I cannot believe what the government is doing — truly shocking — people really need to wake up.”
Or
Before ChatGPT: “any1 no if we still get benfits if ur workin 16 hr askin 4 mate ???”
After ChatGPT: “Does anyone happen to know whether benefits still apply when one works sixteen hours — asking for a friend, of course.”
I very much doubt people understood what an em dash was or why it was being used, they just copy and pasted. Was it an attempt to come across smarter than they are, or was it due to embarrassment at their lack of writing skills, however I suspect it was also laziness.
I’m not knocking the use of AI, how can I? AI, mainly ChatGPT, is a key tool in my blogging toolbox. I mainly use it for brainstorming, and then checking formatting and flow of what I write. I’m very open about my use of AI, and every post has an AI meter at the bottom detailing how much AI was utilised in the post.
The question is why AI uses em dashes so much? Well it seems it’s down to a number of factors.
Firstly, the material used to train the LLMs is generally high-quality writing, literature, journalism, and formal prose. In these types of writing em dashes are frequently used. Secondly, token efficiency, the use of an em dash can be a “hack” for efficiency. Because the em dash (“—“) is typically a single token in the model’s vocabulary, using it often allows the model to condense information into fewer tokens. This is relevant because the cost of using LLMs (e.g., ChatGPT) is calculated by token usage. It’s the LLM way of saving money and energy.
Due to this token efficiency the chances of ChatGPT and the other LLMs toning down the use of the em dash is unlikely.
There’s nothing wrong with the use of the em dash, but the now commonality of it diminishes its use. It used to be almost solely utilised seasoned writers, now it’s seen as a giveaway sign of AI writing.
It’s a shame that the em dash has been hijacked by AI — it deserved better.
Take care,
Mike
Who Wrote This?
Note: 85% me, 15% AI. The usage of the em dash was decided by me and me alone.