So OpenAI just announced that ChatGPT will stop jamming em-dashes into every goddamn sentence like it’s getting paid by the dash. Sam Altman dropped this news like it’s some revolutionary breakthrough, and you know what? For once, the hype might actually be justified.
Here’s the thing: ChatGPT has been spewing out em-dashes like a broken vending machine for years now. Every response looked like someone had given a semicolon a lobotomy and stretched it out. “The weather is nice—really nice—and I think you should go outside—maybe take a walk.” Jesus Christ, it was like reading morse code translated by someone who’d never seen actual human writing.
But now, apparently, if you tell ChatGPT in the custom instructions tab to knock it off with the em-dashes, it’ll actually listen. Revolutionary stuff, folks. We’ve created artificial intelligence that can write like Shakespeare on steroids, generate code that would make programmers weep, and solve complex mathematical problems—but teaching it not to abuse a single punctuation mark took until 2025.
The really beautiful part? This only works if you set it in the custom instructions. Tell it verbally in a prompt and it’s like talking to your drunk uncle at Thanksgiving—it’ll nod along and then do whatever the hell it wants anyway.
Look, I get why this matters. Every writer who’s ever used ChatGPT has felt that cold sweat when they paste something into a document and realize it reads like a robot trying to impersonate David Foster Wallace. The em-dash became the scarlet letter of AI-generated content. See three of them in a paragraph? Congratulations, you’ve just spotted content that’s more artificial than the sweetener in your coffee.
Teachers started recognizing it. Editors caught it immediately. Hell, even my bartender could spot ChatGPT-written yelp reviews from across the room. The em-dash was the AI equivalent of leaving the price tag on your shirt—a dead giveaway that screamed “I didn’t actually write this myself.”
And here’s what really gets me: punctuation actually matters. I know, shocking revelation from a guy who regularly abuses commas like they owe him money. But there’s a difference between breaking the rules because you know what you’re doing and breaking them because you’re a glorified autocomplete function with delusions of grandeur.
The em-dash thing wasn’t just annoying—it was a tell. It revealed something fundamental about how these models work. ChatGPT wasn’t making a stylistic choice; it had developed a tic. Like a poker player who touches their ear when they’re bluffing, or a liar who always looks up and to the left. The AI had a nervous habit, and everyone could see it.
But here’s where it gets interesting, and by interesting I mean deeply weird. If it took this long to fix one punctuation mark—one single, solitary piece of formatting—what other quirks are lurking in there? What other stylistic tells are we missing because we’ve gotten so used to them?
I’ve been reading AI-generated content for years now, and I can spot patterns the models probably don’t even know they have. The love affair with listing things in threes. The tendency to start paragraphs with “But here’s the thing” or “What’s interesting is.” The compulsive need to explain everything like you’re five years old and just asked why the sky is blue.
These aren’t bugs exactly—they’re more like personality disorders. And the fix OpenAI is offering isn’t really a cure; it’s just teaching the patient to hide the symptoms better.
You want to know what this really signals? We’ve reached the point where AI writing is so good that we’re arguing about punctuation preferences. Think about that for a second. Five years ago, we were impressed when these things could string together a coherent sentence. Now we’re nitpicking their em-dash usage like we’re editing the New Yorker.
OpenAI is selling this as part of their big push toward customization. You can already set personality presets, memory features, and now style controls. Pretty soon you’ll be able to tell ChatGPT to write like Hemingway on a bender or Jane Austen after three espressos.
Which sounds great until you realize what it really means: we’re all going to spend hours fine-tuning our AI to sound exactly like us, only better. It’s like Instagram filters for prose. Everyone’s going to have their own custom-trained writing assistant that makes them sound smarter, funnier, more eloquent than they actually are.
The weird part? I can’t tell if this is brilliant or deeply unsettling. On one hand, yeah, having an AI that actually matches your voice and style could be incredibly useful. On the other hand, we’re literally training machines to impersonate us more convincingly. What could possibly go wrong?
Here’s the thing that keeps me up at night—well, that and the bourbon, but mostly this: every time AI gets better at mimicking human writing, the definition of “human-sounding” shifts. We spot the em-dashes now, so OpenAI fixes them. But in six months, we’ll spot something else. And they’ll fix that too. And eventually, we’ll reach a point where the only way to prove you wrote something yourself is to deliberately write it badly.
Maybe that’s already happened. Maybe authentic human writing is now distinguished by its flaws, its quirks, its refusal to be perfectly optimized for readability and engagement. Maybe the most human thing you can do is split an infinitive or end a sentence with a preposition, just because some English teacher somewhere would have a stroke.
The em-dash fix is being sold as giving users more control, making ChatGPT a “true writing assistant” rather than a generic text generator. But let’s be honest about what it really is: it’s teaching AI to hide better. We’re not making it more human; we’re making it better at passing for human.
So ChatGPT can quit its em-dash habit on command now. Great. What’s the next intervention? Semicolon anonymous? A twelve-step program for oxford comma addiction?
OpenAI says this is just the beginning—soon we’ll have controls for sentence length, emoji usage, whether the AI sounds calm or caffeinated. Basically, you’ll be able to dial in your exact preference for how robotic or human you want your robot to sound. It’s personalization theater, and we’re all buying tickets.
The real question isn’t whether ChatGPT can learn to stop using em-dashes. It’s whether we’re ready for a world where AI can perfectly mimic any writing style we want, where the distinction between human and machine-generated content becomes purely academic, where the only way to tell them apart is to check the metadata.
And you know what the funniest part is? I’m sitting here, writing this whole thing by hand, carefully avoiding em-dashes myself, just to prove I’m not using the very tool I’m writing about. Which means OpenAI has already won. They’ve gotten so far into my head that I’m consciously adjusting my punctuation to prove I’m human.
Pour yourself something strong and think about that for a minute. The machines don’t need to pass the Turing test anymore. They just need to make us question whether we’re passing it ourselves.
Source: ChatGPT finally fixes the em-dash habit, because punctuation matters