Why Do AI Models Use So Many Em-Dashes?
Posted2 months agoActiveabout 2 months ago
seangoedecke.comTechstory
calmmixed
Debate
80/100
AI WritingTypographyLanguage Models
Key topics
AI Writing
Typography
Language Models
The article explores why AI models frequently use em-dashes, sparking a discussion on the possible reasons behind this phenomenon and its implications for human writing.
Snapshot generated from the HN discussion
Discussion Activity
Very active discussionFirst comment
43m
Peak period
55
0-6h
Avg / period
8.7
Comment distribution96 data points
Loading chart...
Based on 96 loaded comments
Key moments
- 01Story posted
Nov 2, 2025 at 1:51 AM EST
2 months ago
Step 01 - 02First comment
Nov 2, 2025 at 2:35 AM EST
43m after posting
Step 02 - 03Peak activity
55 comments in 0-6h
Hottest window of the conversation
Step 03 - 04Latest activity
Nov 5, 2025 at 1:51 PM EST
about 2 months ago
Step 04
Generating AI Summary...
Analyzing up to 500 comments to identify key contributors and discussion patterns
ID: 45788327Type: storyLast synced: 11/20/2025, 6:48:47 PM
Want the full context?
Jump to the original sources
Read the primary article or dive into the live Hacker News thread when you're ready.
Your readers won't care about the dashes as long as the texts read like they had human origins and you have something to tell.
That said, yes, keep using them (and using them well!).
I don't understand the purpose of using LLM's to write articles unless someone wants to be the middleman of slop and if that's the case, I'd rather cut middlemans and get slop directly from the AI models, instead of pasting the output of what chatgpt generated, give me the prompt and maybe temperature/other settings if need be to make it more reproducible but the prompt itself could be enough smh
I am not saying you should change your writing style, but at the same time, you have to understand, if someone writes like AI, Chances are that we are too tired of looking too deep into it to find if its written by AI or not, we are tired of it & so you must understand our or anybody's frustration if they call out someone's writing as AI.
For those using AI to write articles/etc. : If you are passionate about something, write about it, write what you want, how you want and you will be proud. But if you use LLM, you will constantly be called upon and frankly, it reduces the purpose of writing.
For code, there is a debate that code is just an means to an end (which is to do stuff like scripts etc.) but there is no end to writing, for what? for more views/etc., there is no point in getting such attention or anything considering it would just be negative attention if I or anyone found AI writing.
Not sure why people use AI text generation for articles etc. Idk.
This is my alt but when I had first started out on HN, I thought my english was fine but then somebody pointed it out and I try to fix my grammar and now its second nature to me writing.
I would be curious to know the reasons as to why people write text stuff with AI in the first place. It doesn't make sense to me since the other side would use their slop to counter your slop, at that point just create a tldr post, why strech an article in more words than unnecessary (I feel like I also write a lot of filler words / yap personally but alright, atleast you know a human is writing this), I don't get the point of writing longer if you aren't even writing it, is it to get SEO or, is the end goal money like all things?
For example, your post could have been just one paragraph and said the same thing. Do you purposefully write so verbosely as a virtue signal of authenticity?
And no, readers can't just ask the LLM to reproduce the same slop, because they don't have the verbose, redundant (there I go again) original source text that it's condensing. And even if they did, they would not bother reading it, because it's tl;dr and full of typos.
Nobody wants to read pages of repetitive human generated slop, either.
PS:
>I thought my english was fine but then somebody pointed it out and I try to fix my grammar and now its second nature to me writing.
Since you asked for somebody to point it out:
Use it's when it's a contraction for "it is" or "it has," and use its when it's a possessive pronoun showing ownership. A helpful trick is to try replacing the word with "it is" or "it has" in the sentence; if it still makes sense, use "it's".
Full disclosure, in case you can't tell: the paragraph above was LLM generated. Did you find it helpful, was it tl;dr, or did you dislike "its" style?
To be really honest, I can understand your view-point even if it conflicts with mine if you aren't being offensive, since personally, I think that there is no point of this offense-defense thing.
Off topic but How's your day going man?
Listen, I will tell you why I write in the way I write, You might say this authenticity but I believe it being honest, I want to give someone access to the thoughts I am thinking the way they come, so I would consider it raw.
Usually its not for them but for me, for knowing how far or backwards I would go in life. I write this because I anticipate reading it in the future but I sure haven't read as much in other places.
HN does feel like a place where someone could write 3 paragraphs to some basic question and still feel accepted or read , to be really honest.
> Nobody wants to read pages of repetitive human generated slop, either.
You raise a good point, I have been a bit selfish in writing these posts, I write it for myself and not for the other person, I thought the other person would appreciate my honesty of typing what I write but the point of it being human slop might make sense too lol
> And no, readers can't just ask the LLM to reproduce the same slop, because they don't have the verbose, redundant (there I go again) original source text that it's condensing. And even if they did, they would not bother reading it, because it's tl;dr and full of typos.
Sorry but either its me but I can't seem to understand what you mean by this? Like, do you mean nobody would bother pasting human slop to get tldr's ?
I don't know what to say since I am not getting what you are trying to tell me here but I am curious for sure.
> Since you asked for somebody to point it out:
I didn't really ask but sure, I will take it. I guess I make less mistakes overall though so that's nice. I am not perfect and I am comfortable knowing that yet I think that my writing could be sharpened, yes. There is no denying in that.
> Full disclosure, in case you can't tell: the paragraph above was LLM generated. Did you find it helpful, was it tl;dr, or did you dislike "its" style?
I would still would've preferred your real message even if it could've been choppy to be really honest.
I feel like If I might not leave an imprint on the world, might as well leave the fingerprint saying I was there and this is a way which helps me feel that way. It is (theraptic?) even to write long sentences, they sooth me. I do it for myself. I wasn't trying to virtue signal though, personally its more that even after anything, I still feel like using LLM's in article formations etc. is just a cheap shortcut to what?? , to me its the fact that I can point to this article and be decently comfortable knowing that I wrote it and not an LLM.
I just can't trust LLM texts that much and the only reason I am giving yours so much is because I would appreciate the opportunity to grow and I am willing to read any criticism you provide me if I can meaningfully work on.
Cmd + Shift + “-“ = —
Let’s spread the word until everyone fancy uses them, and then those who criticize text for coming from LLMs will be ridiculed by our ridiculous skills.
Now I find myself intentionally adding typos and other msitakes, and using less sophisticated language, just to not be accused of using AI.
Ironically, AI writing is too literate. It reads like clunky pastiche to literate readers, but it's still using words and constructions less literate readers haven't seen before.
(Edit: corrected (unintentional) typo)
You can customise the default style over an impressive range. Most people don't, so most AI writing is distilled essence of Failed LinkedIn Marketer, even when that style conflicts hilariously with the content.
line breaks was first; semi-colons was second.
(and yep, I goofed around with both those ;)
So I think you can keep using em-dashes without being associated with LLMs as long as you reserve them for particularly effective/tasteful occasions.
The advent of the generic AI writing style has had one good effect on my own work: making me take an unflinching look at my own laziness in writing. Now I tend to clean things up while at the same time try to inject some personality in order to NOT be dismissed as AI.
One thing I wondered is whether high prestige writing is encoded into the models, but it doesn't seem far fetched that there's various linkages inside the data to say "this kind of thing should be weighted highly."
Thanks to LLMs I learned that using the short binding dash everywhere is incorrect, and I can improve my writing because of it.
I guess in the past if you'd shown me a passage with em dashes I'd say it looks good because I associate it with the New Yorker and Economist, both of which I read. Now I'd be a bit more meh due to LLMs.
The approachable tone of popular print media also preselects for the casual, highly-readable style I suspect users would want from a bot.
I don't have evidence that that's true, but it's what I assume and I'm surprised it's not even mentioned as a possibility.
When I studied author profiling, I built models that could identify specific authors just by how often they used very boring words like 'of' and 'and' with enough text, so I'm assuming that OpenAI plays around with some variables like that which would much harder to humans to spot, but probably uses several layers of watermarking to make it harder to strip, which results in some 'obvious' ones too.
But the watermarking layers is a fascinating idea (and extremely likely to exist), thanks!
https://youtu.be/1d4JOKOpzqU?si=xXDqGEXiawLtWo5e&t=569
It wasn’t just Ev - I can confirm that many of us were typography nuts ;)
Marcin for example - did some really crazy stuff.
https://medium.design/crafting-link-underlines-on-medium-7c0...
Hostile? That’s definitely a take. Curious what you’re thinking there.
Probably the second most annoying thing on the web today is when you click a link that looks interesting but the page you land on almost immediately says you have to do or pay something to actually read the thing the referring page implied. I don't even start reading a Medium article now if I can see that pop-up below - it's just an instinctive reaction to close the tab. I wish people wouldn't link to articles in walled gardens and search engines would remove those articles from their index - or if that's not reliable then exclude entire sites. Those walls break the whole cross-linking model that made the web the success it is and they waste people's time on a global scale.
I recognise that my position may be somewhat hypocritical because I'd rank AI slop as #3 and maybe #1 and #2 are making some kind of attempt to avoid supporting AI slop. But then I'd propose a more draconian solution to that problem as well - one involving punitive penalties for AI companies that scrape others' content without permission to train their models and possibly for anyone else using models that are tainted.
If you feel you’re entitled to everyone else’s labor - I dunno what to tell you.
On the other hand, if you value your own time so little that the only amount you're willing invest in the quality of what you read is $0 - I also don’t know what to tell you.
Either way, I hope you figure it out.
Medium (at least what it is today) tries to bring down the friction of making valuable content available at a reasonable price.
The alternative solutions the web has been to come up with is to take the valuable content and lock it up in hundreds of silos (Substack, etc), leave residual low value content marketing available, and then cover most everything else with a browser melting level of “adtech”
https://en.wikipedia.org/wiki/Doorway_page
You're perfectly entitled to keep your content commercial if you want. Just don't put it in the same place as the freely available material that everyone else was working with and then complain when people find you irritating. Some of us are content to share our own work for free on the web and to enjoy work that is offered freely by others. We're all doing it right now on HN and many of us run non-commercial blogs of our own too. And we made the web an interesting and useful place long before sites like Medium came along and tried to centralise and commercialise it.
I’m also not affiliated with twitter or Elon at all, so not sure what the rest is about.
Not that most websites are any better. My favourites are basically the ones that just show a default "sorry but this content is blocked in your region" text.
Needless to say, I closed the tab. No content is worth dealing with that over.
Sure plenty of other sites do it too but “other people do it” doesn’t mean it’s not hostile nor does it excuse the behavior. Medium is and has always been one of my most hated sites because a lot of tech people post there, a lot of medium links are submitted to HN, yet it’s a horrible place for the reader.
https://news.ycombinator.com/newsguidelines.html
Apple has done it across their systems for ages. Microsoft did it in Word for a long time too.
It was more or less standard on any tool that was geared towards writers long before Medium was a thing.
That explains a lot…
And so AI models are prone to using them because they require less computation than rewriting a sentence.
Never spent too much time thinking about em-dashes. Writers I like probably use them all the time—again, never really thought about it.
There are many other language model artifacts that are genuinely shite and are worth criticizing. Though, come to think of it, they have been getting stamped out with each iteration in model. Used to spend a lot of time trying to get models to refrain from words like "crucial".
What I do find strange is how the latest SOTA models appear to write with contractions by default, which began sometime in the past year. Anthropic models, in particular.
Instead of
"The time has come," the Walrus said,
"To talk of many things:"
... it would be spelled as
— The time has come, — the Walrus said,
— To talk of many things:
I wonder how much of russian language content was in training model.
Actually, they wouldn't have to go and buy these old books: The texts are already available copyright free, due to legislation stating that copyright expires 70 years after the author's death (any book published in the USA before 1923 is also reproducible without adherence to copyright laws), making the full texts of old books much easier to find on the internet!
Also, I wrote a small extension that automatically replaces ChatGPT responses with em dashes with alternative punctuation marks: https://github.com/nckclrk/rm-em-dashes
Yeah, this is me. I've always liked good type and typography. 5 or 6 years ago I added em-dash to my keyboard configs to make typing it in convenient - mostly because I just think it just looks nicer. But lately I don't use it much because... AI.
However, in recent weeks someone accused an HN post of mine as being from a bot, despite the fact I used a plain old hyphen and not an em-dash. There was nothing in the post which seemed AI-like except possibly that hyphen. At the time, I realized that person probably just couldn't tell a hyphen from a real em-dash. So maybe that means I have to not use any dash at all.
Regular computers for human use only support ASCII in US or ISO-5589-1 in EU still to this day, and Unicode reliant East Asian users turn off Unicode input modes before typing English words, leaving the Asian part mostly in pure Unicode and alphanumeric part pure ASCII. So Unicode-ASCII mixed text is just odd by itself. This in turn makes use of em dashes odd.
Same with emojis. LLMs generate Unicode-mapped tokens directly, so they can vocalize any characters within full Unicode ranges. Humans with keyboards(physical or touchscreen) can mostly only produce what's on them.
The truth is ... most people don't read much. So it's not too surprising they think it looks weird if all they read is posts on the internet, where the average writer has never even learned how to make one on the keyboard.
Delve on the other hand, that shit looks weird. That is waaay over-represented.