OpenAI says ChatGPT will now ditch em dashes if users tell it to; em dashes have become telltale signs that supposedly signals text written by AI

OpenAI says ChatGPT will now ditch the em dashes if you tell it to. The telltale sign that supposedly signals text written by AI has popped …

TechCrunch 2025-11-15 Sarah Perez

Context & Ripple Effects

OpenAI’s change treats a widely noticed writing habit as a controllable output preference rather than a fixed model signature. It extends the company’s pattern of reducing conspicuous friction in ChatGPT, including its earlier removal of some content-warning messages.

The move also fits a broader push toward finer-grained response controls, later reflected in settings for tone and formatting choices. Here, the relevant control is not just readability: it affects how easily readers infer that text was AI-generated.

First-order effects

ChatGPT users can request prose without em dashes, making the model’s output better aligned with house styles or personal writing preferences.
A punctuation cue that some readers used to flag likely AI-written text becomes less dependable for ChatGPT outputs when users actively apply the preference.

Second-order effects

Teams reviewing submitted or published copy will have less reason to treat one stylistic habit as a quick proxy for AI use; review shifts toward context, process, and substantive quality.
Competing writing assistants face added pressure to expose similarly granular controls, since output customization increasingly includes avoiding recognizable model habits.

Third-order effects

If AI systems keep smoothing away recognizable stylistic patterns, informal AI detection based on surface cues will weaken faster than it can serve as a reliable governance tool.
The longer-term divide is likely between tools that merely make AI text blend in and workflows that can establish provenance or disclosure through mechanisms beyond the text itself.

The trend: Generative-AI products are moving from generic output toward user-tuned writing that is harder to distinguish from an individual’s preferred style.

Discussion

@sama Sam Altman on x
Small-but-happy win: If you tell ChatGPT not to use em-dashes in your custom instructions, it finally does what it's supposed to do!
@beffjezos @beffjezos on x
em dash should be called ai dash at this point
@grg Greg Kumparak on x
if they could just turn emdashes off by default so I could go back to using them* without judgement that would be great (* way too much, as I have for a thousand years)
@garymarcus Gary Marcus on x
if you can't even get ChatGPT to respect people's em-dash preferences, how the F are you going to solve alignment?
@ibab Igor Babuschkin on x
I'm seeing AI generated posts and replies everywhere now. ChatGPT has a few telltale signs that give away AI generated content. These will soon be post trained away. It's not about AI content being everywhere, it's about the fact that we'll soon lose the ability to tell!
@tunguz Bojan Tunguz on x
Our last defense against AI slop is now down.
@xbenjamminx Ben Jammin on x
This is the biggest improvement in AI that I've seen in the last two years.
@imnotowned Drew on x
Curing cancer right around the corner!
@petetrainor @petetrainor on x
“Help me disguise the Ai-y-ness of my content.”
@thegoodtimeline Ryan Ferris on x
and here i was thinking @OpenAI purposefully kept em-dashes in, as way to detect slop/spam turns out chat just reaaaly loves em-dashes lol
@pocarles Pierre-Olivier Carles on x
The most awaited feature since GPT‑3.
@artbrock34 @artbrock34 on x
Oh, this made-my-day
@benjamingoggin Ben Goggin on x
If it took them this long to get ChatGPT to stop using em dashes, it definitely makes you wonder how long it will really take to address issues like encouraging suicide
@kjovano @kjovano on x
How much pollution will be created to fine-tune technology that isn't going to save people's lives?
@neerajkh_ Neeraj Khandelwal on x
ChatGPT launched the most important feature of 2025: no em-dashes. It works too. I checked. Always verify, then trust. 🤣 [image]
@garymarcus Gary Marcus on x
um, if you can't specify your em-dash preferrences, how the F are you going to solve alignment?
@rahll Reid Southen on x
You guys trust a technology that took them literally years to figure out how to get it to stop using em dashes? Okay then...
@shuaichenchang Shuaichen Chang on x
This reminds me of when Apple announced they launched five new emojis 🙂‍↔️ lol. But honestly, this is great progress. Getting LLMs to follow instructions at the token level is harder than it looks. Now I'm gonna try teaching it to talk without prompting itself into a new task
@biilmann Matt Biilmann on x
AGI!
@beckykeene Becky Keene on x
The ultimate goal is to normalize using em-dashes as humans again so those of us who write with them naturally aren't judged by the AI shamers. #AI
@ralphbrooks Ralph Brooks on x
AI one step closer to being indistinguishable from human writing.
@aspittel Ali Spittel on x
I would like to start an emdash redemption campaign. Sincerely, someone who has been way overusing dashes since before ChatGPT.
@netzkolumnistin Angela Gruber on x
Daily reminder: AI is not omnipotent but extremely situationally dependent.
@menhguin Minh Nhat Nguyen on x
im 70% convinced openai intentionally keeps the emdash as an easy way to watermark text written by ChatGPT.
@drorpoleg Dror Poleg on x
A small step for ChatGPT — a giant leap for writers who love em-dashes.
@kimmaicutler Kim-Mai Cutler on x
Nooooo........ my child was reading the em dashes aloud as minus signs, as in “Bye MINUS have a nice day!”
@sarahcat21 Sarah Catanzaro on x
Sam, this is not small. Huge if true.
@joshuamarch Joshua March on x
Can everyone do this So I can go back to using em-dashes
@marckohlbrugge Marc Köhlbrugge on x
Best thing about this is that we can finally use em-dashes again without being accused of AI writing 🤣
@ben_golub Ben Golub on x
Unfortunately this reduces it to only silver medal level on the IMO due to the strain of pushing so hard against its nature.
@yanatweets Yana Welinder on x
Only took 3 years to teach AI to stop using em dashes. Now the real test: how long until humans stop treating every em dash as an AI tell?
@dkundel Dominik Kundel on x
As a user of em-dashes I'm way too invested in AI writing less with em-dashes 😄
@chrismaddern Chris Maddern on x
biggest — chatgpt — upgrade — ever
@gwintrob Gordon Wintrob on x
All the personalization you need: When writing text, avoid em-dashes and other common “AI slop” patterns.
@rhyssullivan Rhys on x
feel free to hit me up next time [image]
@lateinteraction Omar Khattab on x
instruction following is hard
@theo @theo on x
Fun fact: there are 23 em-dashes in the GPT 5.1 announcement post [image]
@ai_for_success AshutoshShrivastava on x
@sama Finally it's over... [image]
@koltregaskes @koltregaskes on x
@sama AGI has been achieved. 😜
@chipro Chip Huyen on x
Sam!!! [image]
@abby621 Abby Stylianou on x
I went through my @CVPR submission last night to remove any em dashes I had written because I didn't want reviewers to think I had used LLMs for my writing. Look forward to a day when I can use em-dashes again without fear.
@simonhedlin Simon Hedlin on x
@sama Great! Hopefully this means that those of us who like em-dashes (the most versatile punctuation mark) eventually can start using them again without being falsely accused of relying on unedited AI writing.
@egoraptor Arin Hanson on x
I find what's being unsaid here disturbing: the reason removing em-dashes from ChatGPT responses is a “win” that is worthy of an announcement is that em-dashes are primarily used to sniff out if a published work was written by ChatGPT. What purpose does this serve except to more
@brian_lovin Brian Lovin on x
AGI has been achieved internally
@harrymccracken.com Harry McCracken on bluesky
Excessive use of em dashes would be about #25 on my list of telltale signs something was written by AI. And it's far from the most annoying one. — (My personal #1 annoying telltale sign is endless use of “It wasn't just x, it was also y”-type constructions.) techcrunch.com/202…
@pileofgarbage.net @pileofgarbage.net on bluesky
This is fucking pathetic and at stark odds with the magical AGI fantasy perpetuated by Altman and his ilk. Focussing on what is ultimately aesthetics to the extent that you're micro-managing the entire gen AI model is literally the furthest possible point from AGI. arstechnica.c…
@rjcc Richard Lawler on bluesky
There are several pretty easy tells for ai writing, em dash overuse isn't really one of them. — One of the reasons they're easy tells is because people who are using chatbots to write, also aren't good at editing. [embedded post]
@nitishpahwa.com Nitish Pahwa on bluesky
the thing about chatbots is that their predictive engines will engender certain patterns more often than not. the em dash factor is a meme; there are tells that go far beyond that [embedded post]
@thealexknapp Alex Knapp on bluesky
As an enthusiastic em-dasher, I really resent that this has come to be a sign of AI-generated copy. [embedded post]
@scriban @scriban on bluesky
This is all—and I say this with respect—so very stupid. [embedded post]

Chronicles