By skyforbes 
OpenAI says it has finally solved ChatGPT’s em‑dash problem.
Last week, Sam Altman posted on X that if you tell ChatGPT not to use em dashes, it will follow your instructions.
That announcement eases a frustration many users have had for a long time, because despite carefully written instructions and updates to memories, ChatGPT just wasn’t listening.
The panic about em dashes took off online when social media latched onto the idea that seeing one meant the text was written by ChatGPT.
The result was a small moral panic.
People started editing perfectly normal punctuation out of their work just so they wouldn’t look “suspicious,” which was frustrating (and a bit humiliating) for writers who had always loved the long dash and used it well before AI entered the room.
If you’re a writer, you know what you like.
Imagine if ChatGPT suddenly refused to use paragraph breaks. Every thought would run straight into the next, creating walls of text that make readers want to run away. You could write the most brilliant insights, but nobody would read them because the format itself is hostile.
Our eyes need natural pauses. We need visual breathing spaces that signal when one idea ends and another begins. Without them, even simple content becomes exhausting to process.
ChatGPT overused the em dash so heavily that many writers had to ban it entirely to make their writing sound natural.
The upside of OpenAI’s change is that if you like the em dash, you can keep it. If you don’t, you can finally turn it off. It’s no longer all‑or‑nothing. Instead, it’s a stylistic choice again.
But the more interesting question is why ChatGPT was so intent on the em dash in the first place.
So what is the EM dash used for?
Look closely at how ChatGPT uses it and you’ll see the em dash for the classic connector that it is.
This isn’t some quirk that emerged with AI.
The em dash gets its name from traditional typography, where it’s the width of the letter M. Writers have relied on it since the printing press era to signal interruptions, add emphasis, and create rhythm in prose.
Em dashes, en dashes, these are all ways of stitching ideas together while keeping readers inside the same thought stream.
They also mirror how we think and speak when we’re exploring something out loud.
It’s often easier to lean on a connector, whether it’s a dash, a “so,” a “but” or an “uhmm” it’s a chance to stop, zoom out, and decide what you actually want to say next.
That’s what makes the em dash such a natural fit for large language models, because it lets them keep going at pace.
It smooths over gaps, glues clauses together, and gives the illusion of flow, even when the underlying ideas aren’t fully developed.
The fix OpenAI just shipped is useful. But the real opportunity for humans is to notice where we’ve been accepting the model’s default connectors and to decide, sentence by sentence, which ones we actually want to keep.
And this won’t be the last LLM quirk we see. It’s just one that garnered the most attention (for now).