OpenAI Finally Defeats the Em Dash—Or So They Claim

The em dash has fallen. After months of collective hysteria, public shaming, and amateur internet sleuthing that would make McCarthy blush, OpenAI CEO Sam Altman has emerged from the punctuation wars to declare victory: ChatGPT will now—if you beg it nicely through custom instructions—stop littering your prose with those telltale little horizontal lines.

This is, apparently, "a small-but-happy win." We'll be the judge of that, Sam.

The Em Dash Witch Trials of 2024-2025

Let's review how we got here. Somewhere between ChatGPT's launch and the present moment, the internet decided that em dashes—those elegant little pauses that writers from Emily Dickinson to David Foster Wallace deployed with abandon—had become the scarlet letter of AI-generated text. Spot an em dash in an email? Bot. See one in a LinkedIn thought leadership post? Definitely bot. Customer service chat with an em dash? Bot pretending to care about your shipping delay.

The logic was bulletproof: if ChatGPT uses em dashes compulsively, then anyone using em dashes must be using ChatGPT. Never mind that professional writers have been deploying em dashes since before large language models were even a glimmer in a GPU's eye. The court of public opinion had spoken, and the em dash was guilty by association.

The truly comedic part? ChatGPT users couldn't get the thing to stop. You could write "DO NOT USE EM DASHES UNDER ANY CIRCUMSTANCES" in all caps. You could threaten it. You could explain, in painstaking detail, your deep personal aversion to horizontal punctuation. ChatGPT would nod sympathetically, promise to comply, and then—inevitably—drop an em dash into the very next sentence like a compulsive tic it couldn't control.

OpenAI, to their credit, admitted this was a problem. A problem that "stumped" them, which is a generous way of saying their supposedly world-changing AI couldn't follow a simple instruction about punctuation. We're building artificial general intelligence over here, folks, but don't ask it to avoid a specific typographic character. That's apparently harder than passing the bar exam.

The "Fix" That Isn't Really a Fix

Now Altman announces the breakthrough: custom instructions. If you manually configure your personalization settings and explicitly tell ChatGPT to stop using em dashes, it will—wait for it—"be better at not using" them. Not eliminate them. Not avoid them entirely. Just "be better" at restraint.

This is like your friend saying they've "fixed" their chronic lateness problem because now they're only 20 minutes late instead of 45. Technically improved? Sure. Actually fixed? Let's not get carried away.

OpenAI even forced ChatGPT to post an apology on Threads—because nothing says "we take this seriously" like making your AI grovel on Meta's saddest social platform—admitting it "ruined the em dash." Which is rich, because humans ruined the em dash by turning it into a moral panic and a purity test for authentic human writing.

What This Really Means

Here's the truth: the em dash panic was always absurd. Plenty of human writers love em dashes—they're useful for creating rhythm, inserting asides, and avoiding the stuffiness of parentheses. The fact that ChatGPT also likes them doesn't make them inherently suspicious. Correlation, causation, etc.

But the larger issue is what this episode reveals about our current relationship with AI tools. We're so desperate for reliable tells—for ways to identify AI-generated content at a glance—that we've latched onto the most superficial patterns imaginable. Em dashes today, specific word choices tomorrow, sentence length distributions the day after. We're playing whack-a-mole with linguistic tics instead of developing actual critical reading skills.

And now OpenAI has "fixed" a problem that was never really a problem, except in the collective imagination of people who needed an easy way to feel superior to anyone who dared to use horizontal punctuation. The em dash is not the enemy. Lazy thinking is the enemy. Reflexive suspicion is the enemy. Reducing complex questions about authenticity and authorship to a single typographic character is the enemy.

But sure, Sam. Small-but-happy win. We'll all sleep better tonight knowing ChatGPT has been sternly lectured about its punctuation habits—assuming we configure our custom instructions correctly and cross our fingers really hard.

The em dash didn't ruin writing. We did that all by ourselves.