Don’t Cry for Me Em Dash, the Truth Is I Never Left You

Worries about authenticity from an AI optimist

I recently received an email that struck me as a little off, though it took me a few moments to put my finger on why. The message ran 392 words and contained 13 em dashes, which works out to one em dash for every 30 words. Fortunately, I had just finished rewatching The Terminator on Netflix, which handed me the answer. That was not a human-written email.

AI is changing how we write, market, communicate, and produce thought leadership. That’s not necessarily a bad thing. For the record, I’m an AI optimist. In fact, as I’m writing this, Codex is checking a model I had Claude Code build.

Still, I’m worried. Much of what’s going on is downright silly. More importantly, it begs the question: When AI functionally builds the message, does the human behind it even matter? Who’s doing the communicating, anyway?

This has significant implications for relationship-driven businesses like ours, as well as for thought leadership and communications more broadly. Radical homogeneity in the texture and style of AI messaging could trigger a form of “ad blindness,” which is a phenomenon in marketing where users start ignoring content (usually ads) due to overexposure or irrelevance.

The Ship of Theseus

The Ship of Theseus is a thought experiment about the nature of identity. One by one, the planks on Theseus’ ship rot and are replaced. Eventually no plank is original. Does this mean it’s no longer the same ship?

Now apply this paradox to AI-assisted writing. Suppose you upload a draft to AI and it rewrites the introduction, retools the structure, trims the copy, swaps in all new adjectives and adverbs, and uses heavy jargon. What’s left of you in the finished product? At what point has the original wood been replaced entirely?

I’m all for digital assistance. AI is extraordinary at editorial chores, and a writer refusing to use it is like a mathematician refusing to use a calculator. The question is how to use this tool while keeping enough of yourself in the final work that the person on the other end still feels like they’re interacting with you.

Taking aim at robotic patterns

One reason AI comes across as inauthentic (and, therefore, irrelevant) is its over-reliance on certain language patterns. Here are three I keep noticing.

Em dashes

I used to love em dashes. These horizontally-advantaged punctuation marks let you slip in a quick aside or a witty sentence ending without breaking your sentence in two. I’d toss about four to five into a paper of this length—sparingly, but just enough to give my writing some texture.

AI does not use em dashes sparingly. AI deploys them like a first-generation Hummer guzzled gas. With most AI-generated content, every second or third sentence has an em dash with forced parenthetical interruptions. The ratio of 30 words to one em dash that I mentioned before may be on the extreme end. But ratios of 50:1 or 60:1 are everywhere now.

The problem isn’t so much about em dashes per se as it is about a generic writing style that bears too little of the writer. Tomorrow the models might have em dashes trained out of them and, instead, use semicolons every other sentence.

Cliché phrases

Months ago, I found myself reading articles and social media posts and thinking, “Wow, this was really well written!” I read more and more of these, continuing to feel like some really healthy writing nutrients had found their way into the water supply. And then my pattern matching kicked into gear.

The problem was the articles were all using the same virtually identical phrasing. The phrase that prompted me to connect the dots was “quietly happening.”

Something is always quietly happening in AI-generated prose. The shift is quietly underway. The transformation is quietly accelerating. The change is quietly reshaping the industry. If everything is quiet, and every article is alerting us to something that is quietly happening, then nothing is quiet and the phrasing is frustratingly loud.

A second cliche is the drumroll. It comes under taxonomies, such as “What this really reveals is,” “What this actually means is,” and “What’s really going on here is.” These phrases promise depth, and I’ll live with them if that’s what truly follows. But I’m afraid drumroll fatigue is starting to set in.

Then there’s the one I call the oracle. Its family of phraseology includes, “We’re seeing a shift,” “The landscape is evolving,” and “The market is changing.” You don’t say!

The dramatic triple tap

There’s a paragraph structure I keep encountering that I now associate almost exclusively with AI-generated content. Three words, each given its own paragraph, separated by periods.

Like this:

Clarity. Impact. Alignment.

Or:

Trust. Consistency. Results.

I asked an AI tool what this device is called (since I had rarely encountered it prior to some date in 2023), and it labeled it the “dramatic triple tap.” There was a time when this construction felt original. Now that I see it in every third article, it’s a lot less exciting. Nothing signals “AI wrote this” faster than the dramatic triple tap.

How to use AI without losing yourself

Make no mistake. The productivity gain of AI is real, and only increasing. The last time I was this excited about tech was when I first encountered the internet—dial-up modem and all. (I’ve now hit my self-imposed two em-dash maximum.)

That said, we’re still working out how to embrace AI in this narrow use case of messaging and communication.

Think of AI’s writing capabilities as a baseball glove. It’s always better than not having a glove. But out of the box, it’s stiff and generic. You haven’t broken it in yet, haven’t shaped the pocket, haven’t polished it, and haven’t worn it until the leather knows your hand. Until you have, it’s still more like you’re borrowing your buddy’s glove, and it’ll show.

Or, think of Barney Ross’ plane in The Expendables. You couldn’t separate that asset from his identity; it was an extension of him. That’s the goal with AI. It should become an extension of how you think, not a replacement for it.

Two practical suggestions.

First, slow down. If all you’re doing is feeding raw input into AI and forwarding the output, you’re asking to be disintermediated. There’s not enough value-add. The answer has to be more than, “I pressed the button.”

Second, use AI in ways that make you think more, not less. The best uses involve more critical engagement on my part, not less: stress-testing my arguments, surfacing counterpoints, finding holes in my logic, asking it to argue the other side.

Treat AI as a teammate, and make sure you’re the team captain.

What’s My Style?

Let me talk about my style for a moment, with the premise that writing is interesting because there’s a plurality of styles and not just one homogenous voice.

Personally, I don’t like opening with punchy one-sentence hooks, because they feel gimmicky to me. I write in what I hope is a helpful and engaging way because I want to share something I find interesting. I want my sentences to breathe, my transitions to feel natural, and above all my personality to come through. I’d rather offer direct observations than lean on cliches or clever rhythms where the words devolve into platitudes versus concrete points.

As much as I enjoyed em dashes, now I’m terrified of them. Colons and semicolons are my new jam. I feel like I’ve gone from being a dessert guy to a salty and savory guy with this transition. I periodically miss my turtle pecan cluster, but the chips and guac taste pretty good too.

Wrap-Up

AI has its place. But your je ne sais quoi is the part of you AI cannot replicate. It’s in how you think, what you find interesting, the unusual connections you make, the things you care about, and the things that bother you. By wrangling control, you can bring those attributes to your writing while taking advantage of the quality and productivity benefits AI has to offer.

Replace the rotted planks. Build a better ship. Just retain enough of the original character that anyone can see the results as authentic.

Disclaimer:

I promise I didn’t wake up on the wrong side of the bed when writing this. I’m good with a few extra em dashes, hearing that something is quietly happening beneath the surface, and learning that the landscape is evolving.