FIR #481: The Em Dash Panic — AI, Writing, and Misguided Assumptions

Sep 24, 2025

Ask episode

Chapters

Transcript

Episode notes

In this short midweek episode, Neville and Shel dive into one of the hottest debates in communication today: what happens to tone and authenticity when artificial intelligence steps into the writing process? From the surprisingly heated arguments over the humble em-dash to fresh research on AI’s “stylometric fingerprints,” we explore whether polished AI-assisted prose risks losing the human voice that builds trust. Along the way, we look at how publishers like Business Insider are normalizing AI for first drafts, how communicators are redefining authenticity, and how Shel used AI to turn years of blog posts into a forthcoming book.

Links from this episode:

The next monthly, long-form episode of FIR will drop on Monday, September 29.

We host a Communicators Zoom Chat most Thursdays at 1 p.m. ET. To obtain the credentials needed to participate, contact Shel or Neville directly, request them in our Facebook group, or email fircomments@gmail.com.

Special thanks to Jay Moonah for the opening and closing music.

You can find the stories from which Shel’s FIR content is selected at Shel’s Link Blog. Shel has started a metaverse-focused Flipboard magazine. You can catch up with both co-hosts on [Neville’s blog](https://www.nevillehobson.io/) and [Shel’s blog](https://holtz.com/blog/).

Disclaimer: The opinions expressed in this podcast are Shel’s and Neville’s and do not reflect the views of their employers and/or clients.

Raw Transcript

Shel Holtz (00:01)
Hi everybody and welcome to episode number 481 of Four Immediate Release. I’m Shel Holtz.

@nevillehobson (00:08)
And I’m Neville Hobson. In this episode of For Immediate Release, we’re going to explore the question of tone and authenticity when artificial intelligence becomes part of the writing process. That seems to be a bit of a hot topic these days from what I see online. AI tools don’t just generate text. They also polish, rewrite, and shift tone to make communication sound warmer, more professional, or more concise. But what happens to authentic voice when AI smooths the edges?

Do we risk losing individuality, nuance and trust if everything starts to sound the same? We’ll talk about that right after this.

It’s a debate playing out among communicators. This year, the humble M-dash has become a flashpoint. Some insist that overusing M-dashes is a dead giveaway of AI altered text. Others push back saying that’s nonsense and unfairly stigmatizes a perfectly good mark of punctuation. Washington Post ran a feature in April with the headline, Some people think AI writing has a tell. The M-dash writers disagree. Then in August, Brian Phillips wrote a lyrical defense in the ringer.

pleading, stop AI shaming our precious kindly M-Dashes, please. And McSweeney’s even joined him as satire, publishing the M-Dash response to the AI allegations written from the dashes own point of view. That is really, really very amusing, worth a read. The fact that such debates exists highlights how sensitive people are to the signals of authenticity in writing. Fresh research in 2025 suggests this is more than speculation.

Some recent studies show that AI leaves stylometric fingerprints in writing that can be detected, raising questions about authorship and voice. A stylometric fingerprint is the unique combination of statistical linguistic features within a piece of text that identifies its author much like a human fingerprint. AI can make writing clearer and more polished but risks homogenizing style and raising ethical questions. Beyond academia, commentators argue that polished words without voice

risk-leaving communication hollow. And while researchers are busy analyzing stylometry and psycholinguistics, communicators are having a very different kind of debate about punctuation. So while academics study the fingerprints AI leaves on writing, the popular imagination has latched onto something much simpler, the punctuation choices we make. The M-debate may be tongue in cheek, but it speaks to a serious point.

How sensitive we’ve become to the signals of authenticity in text right down to a single line on the page. For communicators, the challenge is not whether to use AI, that ship has sailed, but how to preserve authenticity when tone shifting tools are in the mix. The call to action is to define what authenticity means in your context, decide which writing tasks AI should support, and ensure human voice and accountability remain front and center.

In the end, and authenticity aren’t about perfectly polished words. They’re about whether people believe there’s a human voice and accountability behind the message. Your thoughts, Shale?

Shel Holtz (03:19)
I have a lot of thoughts on this, ⁓ beginning with, well, at least it’s not the Oxford comma, because that’s a source of debate without artificial intelligence. The dash has found its way into AI outputs because of the AI inputs. The training sets include this massive corpus of writing that has been scraped from the web.

@nevillehobson (03:28)
Don’t start on that.

Shel Holtz (03:49)
that includes dashes. This is what it learned from. It saw a lot of ⁓ dashes used in writing, and that’s a pattern that it recognizes. It recognizes where they’re used and implements it in the outputs it creates. It did not think to itself, you know, I haven’t seen many dashes in my training set. That’s a shame. I’m going to start using more of those. That’s absurd.

When I was in college, I had a part-time job setting type. Yeah, I’m old enough that I actually set type. And I remember learning when to use the ⁓ dash based on the copy that I was transcribing into typeset. And there were a lot of them, even back then, when most people were still working on typewriters. So I think this notion that it’s a tell is ridiculous.

It’s, as you quoted somebody saying, a perfectly serviceable bit of punctuation. But let’s go beyond this notion of punctuation. mean, leave it to communicators where we have some really weighty issues to deal with that we’re going to spend most of our time talking about punctuation. I think this is one of the reasons leaders don’t take communicators very seriously. They’re thinking about business decisions and we’re thinking about

⁓ letting and kerning and things like that. We really need to get more focused on business outcomes and how communication and the use of AI contributes to that. This is what the Business Insider has done. Business Insider, I don’t know if you saw this, it’s a very recent announcement, have ⁓ officially given permission to their journalists to use AI to write the first drafts of their articles. They were already…

@nevillehobson (05:44)
Yeah.

Shel Holtz (05:47)
allowed to use AI for research, but not for any of the drafts they produced. Now they can produce the first draft. Now, why was that decision made? I’m not privy to what was going on in the mind of the president of Business Insider who made this decision and communicated it through internal memo to her staff. But I have to believe a couple of things are in play. First of all, journalism is in financial difficulties and you need

fewer people to crank out more stories. And if you can get it done faster by having AI generate a first draft, you then go in and fact check and clean up and apply your own writing to, I’ve done this, I’ve done first drafts in AI. And by the time I’m done rewriting, it’s a completely different piece. Still took me about an hour and a half less than it would have if I had had to sit there and write the first draft. I don’t do that for every kind of

article or other material that I need to produce, but on some things it just makes sense and it makes life easier. But the other reason I think Business Insider decided to go down this road is because AI is getting better at producing these kinds of drafts and it’s going to continue to get better. In the world of business, and I’m sure you’ve had this experience Neville, I know I have, is outside of the world of

communication when you are dealing with people in other parts of the business. And I don’t mean this in any sort of pejorative way. This is not an insult. These people are brilliant when it comes to the areas of specialization that are the focus of their jobs. But they can’t write their way out of a wet paper bag. They’re terrible, terrible writers. And if using AI can help them write a memo, write an email, write a report, write an article,

better than they could have on their own, I think we’re at a point where that’s fine. And as I say, it’s going to continue to improve. If AI can generate a good first draft now, how many months or years before it can generate the good final draft? I heard Casey Newton talking about this on Hartford saying, yeah, maybe ⁓ there’s not going to be work for us journalists anymore when the AI can write all the drafts.

don’t think that’s necessarily the case because the research still involves talking to people and that sort of thing. the writing with AI is getting to the point where I think this is a very transitive conversation that we’re having about authenticity and human writing. think AI is fully capable of doing this in a lot of circumstances and we need to stop our hand wringing about it and figure out how to do it well.

So that it produces the best results we can possibly deliver for our clients and our companies

@nevillehobson (08:50)
I agree. But we’re talking about human beings here who don’t exhibit some of that sometimes. I think the key thing to me is this kind of phrase I found last on my own mind, AI assisted writing. That’s actually pretty accurate phrase to describe in a sense what we’re talking about here. The issue of tone and authenticity is really the point that I’m keen to explore, where

voices more than grammar, for instance, AI can clean up text, sure, it can improve text. And you could argue that’s a subjective way, phrase, improve. Others might not see it as improvement. And therein, you get into rabbit holes, without question, which is where we’re at in a lot of this, I think. But AI can clean up text. But I think authenticity demonstrates itself through word choices, rhythm.

even the imperfections. I did something the other day that ChatGPT5 assisted me with, and I thought about this when I mentioned that. Did I assist it or did it assist? No, it assisted me because I asked it to do things. And I noted, kind of taken aback slightly, that on the first draft of what it did, it told me what it was going to do. And it said, I’m going to give this to you in your tone of voice.

And it did. And I read this, goodness, I could have written this myself. And that did take me back a bit. Maybe I’d not noticed that before. But that example also prompted a lot of my thinking into leading up to this discussion today. Because tone ⁓ and authenticity or rather tone, yeah, I mean, it is authenticity.

even though the debate is still there about what defines authenticity, I’ve had discussion that with lot of people is what do mean authentic? What does that actually mean? And it means different things to different people. I think this discussion or debate or whatever we might call describe it is likely to be one of those never ending ones, the dash, you mentioned the Oxford comma, that’s been around for decades, it’s still a major issue for some people. But being clear on this,

And for the reasons I think are quite clear, you mentioned, you know, the large language models, they scrape all this stuff off the off the internet, and it’s got m dashes in, it’s got all this stuff. No one really noticed that until recently. Most people don’t even think about that. I see writing and I even used to use this where I didn’t do m dashes, because that’s not in my background of writing. I do, for instance, on a keyboard.

using WordPerfect back in the day, I might do double hyphen that would convert that into an M dash. I never thought of that even, I didn’t even think about, hey, that’s an M dash, I didn’t even notice. So I don’t like M dashes, actually, the formal way people use them, or the historical way which I see AI doing the same, you got a word, there’s another word, and there’s an M dash in between, and then they’re touching the word. I don’t like it. I don’t like the look of it. So I don’t use M dashes, I use N dashes mostly.

when I do my writing. So this looks neater in my mind and I don’t care about it’s some ⁓ grammar geek is looking at that thing here that alters the meaning of what you’ve written. No, it doesn’t.

Shel Holtz (12:19)
Yeah, and

just to point out when I was doing typesetting, and I don’t remember what the rules were because I was 19 at the time, but there were definitely rules about the M dash is used in these conditions and the M dash in those conditions. I don’t remember what they were, but they existed.

@nevillehobson (12:22)
Sure.

I know,

of course there are rules, of course there are rules. So, but today in 2025, we’re looking at something that is, I guess, out of out of anyone’s control. Now, this this thing will evolve the language. I’m pretty certain of that, particularly. And typically we talk about the English language. The same thing applies in other languages, but English language as the world’s most spoken language.

meaning not native, it doesn’t matter, second language, third language doesn’t matter. It’s different everywhere. And so the rules are shifting. So, you know, I look at ⁓ the economist guides, the APs guide and all this, and indeed, I’ve got a copy of the AP guide from 15 years ago. don’t think anyone follows that now. It’s different. It’s different what you see now. So the ⁓ the web we’re at now is ⁓ beyond ⁓

can you say the the use of AI? I mean, it’s a given, as I mentioned in my intro, whether to use AI that ship has sailed long time ago. But we’re talking about preserving authenticity that exercises a lot of people in right, hence the criticism. I did read a piece by ⁓ I’ve forgotten who it was. It might have been Anne Handley, I think might have been the grammar girl actually talking about this recently. That’s right, right, exactly.

Shel Holtz (14:00)
That would be Minyung Fogarty.

@nevillehobson (14:04)
that ⁓ was pretty ⁓ bullish about the use of this in ways that we shouldn’t be wasting energy and time talking about this. Yet there are many people, because I see them talking about it online and social networks, who are convinced that an end dash anywhere they see an AI has written it. And I don’t think we’re to be able to get rid of that anytime soon. In which case, my advice is just do not worry about that. No one, know, right, ignore them.

Shel Holtz (14:28)
Yeah, ignore them. Ignore them.

I recently finished the first draft of a book. people who have been following writing. Oh, no, I’ve finished lots and lots of books reading, no, writing. This is a book on employee communication. And it is based on a series of 28 blog posts I wrote based on a framework I developed about 10 or 11 years ago.

@nevillehobson (14:39)
Do you mean reading it or writing it? Reading it or writing it? Writing it.

Right, okay.

Shel Holtz (14:58)
to aid me in the consulting work I was doing at the time. I wrote all these blog posts before AI was available to help you write. So they’re all in my voice, but they’re blog posts. They’re not book chapters. So I didn’t have the time between my full-time gig and FIR and a couple of volunteer activities that I have to sit and rewrite everything. I started, I tried, and it was just way too time consuming.

And I was listening to Chris Penn talk about how he did a book that Amazon could not challenge him that it was his, even though he leaned heavily on AI for the work. I didn’t do what he did because I already had a first draft with those blog posts, but I adapted his concept. So here’s what I did. I first created a very hefty document that contained a lot of what I have written.

⁓ I still had PDFs of a couple of the books I had written before. I had all of my blog posts, ⁓ many, many articles, this type of thing, even transcripts from some FIR episodes. And I put them in a PDF. And then I created a document ⁓ that said, these are approaches I take to writing. These are words and phrases that I use. I do a lot of parenthetical statements.

And here are things that AI tends to do in writing that I never do and don’t do any of this. And then I created a gem over on Gemini called Write Like Shell. And I loaded both of those documents into it. This was just for this book. And then chapter by chapter, and I started off by giving it the instruction that I’ve got 28 blog posts. I need to turn them into chapters of a business book.

I also need to update case studies because many of these are 11, 12, 13 years old and no longer relevant. And I need to come up with the kinds of elements that you see recurring in chapters of a business book. So I’m going to give you my blog post one at a time using these documents that I have shared, my corpus of my writing and my instructions on my writing and AI writing and how to apply that. Rewrite this post.

as a book chapter. And I was knocked out at how well it did. I still needed to spend an hour on editing and rewriting and making revisions. But I ended up doing that for 28 chapters. Is that an AI written book or is that a shell written book? I would argue that that’s my book. It based what it did on my writing. It did its new version based on how I write. And then I went back and made sure it was my writing.

@nevillehobson (17:48)
Right.

Shel Holtz (17:55)
So, you know, when we talk about using AI to write, the question is, what is your process? Are you saying write this and then you go publish it? I would never do that, but I think there are a lot of people who do.

@nevillehobson (18:03)
Right. No, no, ⁓

I agree with you. In fact, what you described is to my mind fits precisely the the label of AI assisted writing. That in my mind, what you’ve just explained is what’s the difference between what you did and between what you might have done if you’d hired a human being assistant to help you do this? What’s the difference? The AI is faster.

Is it accurate? Well, you only you will know that and you’re going to review it all anyway. And this to me is is is totally fine. And you’ve got no you would have no alarm about well, should I kind of write this chunky paragraph explaining that I’ve done it like this? No. Why? Why would you do that? Because similar tools have existed until now that were nowhere near as good as this. I mean, I think if some of the grammar assistance you

get from even in Microsoft Word back in the day, we’re Grammarly now that’s evolving that uses AI. And it’s, ⁓ you know, we got all these tools. Right, you’ve got all this there, which kind of neatly circles back to the main issue here, which is tone and authenticity, how do you still bring that along? And I think that is something where the the the user, the prompter, the communicator, whoever who is

Shel Holtz (19:01)
Yep. Hemingway is another one.

@nevillehobson (19:25)
engaging with the AI is his or her absolute responsibility to set the parameters for the AI in that context, to educate the AI on you. I’ve done that a lot. spent a lot of time on chat GPT. And that was one of my big bug bears, by the way, about chat GPT five, it seemed to have forgotten half of what I told chat GPT four about. So I’ve redone all that with chat GPT five. And I can see that in how it

ask me questions, how it prompts me and how responsible I ask it to do stuff like that. So that to me is is one way in which you can be confident yourself that the tone and authenticity of your AI assisted writing is okay. And that really is really what it comes out is not about creating a kind of a corporate manual for everyone has to follow. And I think it’s also to do with the comfort levels you have in your perception of that.

piece of writing or reading that someone has written that may have been assisted by AI may not more than often than not, it’s likely to have been assisted by AI. What is wrong with that? Well, a of people find something wrong with that nevertheless. So go back to right.

Shel Holtz (20:38)
Yep. And again, I think that’s that transitory issue that I think we face. I think in three or

four years, no one’s going to be talking about

@nevillehobson (20:45)
No, I’m sure you’re right, Shell. I mean, I think now we’re seeing things emerging in these academic papers. One of the examples I mentioned about stylometric fingerprints. think we’ve hopefully I think we’ve passed now through the phase of people saying, hey, I’ve got this software tool that could tell instantly if something’s written or not. No one cares about that unless you are being deceitful and seriously trying to game something. You ain’t going to succeed with that even. So things have changed. Things are evolving real fast. And people’s

if you like, expectations are evolving too, particularly the younger you are. And according to one of the papers I read, younger audiences are comfortable generally speaking with AI polish, but care deeply about whether a message feels authentic. And that’s what we need to be paying attention to. So help in defining authenticity is one job communicators could take on. We’ve got through the training, yes, we know, here’s how you do this with this LLM.

Here’s what you could do with that. Now let’s pay attention to the tone. Are you paying attention to that? So I think the kind of road is shifting in a slightly different direction than otherwise we might expect because of this. And the academic research seems to be backing this up as well. Things are changing fast, it seems to me, and you’ve got to be on the case.

Shel Holtz (22:03)
Yeah, and by the way, I have sent this first draft of my book to a group of thought leaders in the employee communication space. I asked them for their input. You know, what needs to be fixed? Is it terrible or should I proceed with this? And if you like it, please write a testimonial. ⁓ And so far, not one person has come back to me and said, I could tell you used AI in this. Not one.

So, you know, if you do this right, if you use it as a tool and do it right and come up with a workable process, I guarantee you all those people who say, can tell when it was written by AI, they can tell when it was written by AI by somebody who didn’t know what they were doing when they prompted the AI to write something. That’s what they can tell.

@nevillehobson (22:32)
Maybe we should ask them.

Yeah,

that’s the watchword, I think. And so, by the way, I see this year, I think there’s been at least six books published about internal communication. So everyone’s on a book bandwagon. I hope yours is gonna be slightly different, because they’re all saying the same thing about strategy, what’s important, how to use AI. Yours will be different, right?

Shel Holtz (23:13)
Mine does, I think it barely references AI. No, mine is about a framework for how to, if you read those other books and you go, okay, I understand the models and I understand how to develop a strategic plan, but I still don’t know how to apply this on a day-to-day basis. The framework is how you do that. One the very first things I say in the book is you’re not going to learn the models of internal communication in this book.

@nevillehobson (23:30)
Perfect.

Shel Holtz (23:42)
Go back and read something fundamental, then come back and read this to figure out what to do with all of those models that you’ve learned about. So yeah, it’s different.

@nevillehobson (23:51)
Cool. Excellent.

Shel Holtz (23:54)
So hope people like it. ⁓ If I actually, I don’t know how I’m gonna publish it yet. I don’t know if I’m gonna self publish this or look for a publisher. I’m waiting for the feedback from the thought leaders that I have sent it to who have agreed to read it. ⁓ But it wouldn’t be done. And this is the point. It wouldn’t be done if I didn’t have AI to help. It would still be sitting there as 28 blog posts. So, you know, criticize it all you like. ⁓ It works.

And that’ll be a 30 for this episode of Four Immediate Release.

The post FIR #481: The Em Dash Panic — AI, Writing, and Misguided Assumptions appeared first on FIR Podcast Network.