The Em Dash isn't an AI Fingerprint. You just Don't Understand Punctuation.
Most people couldn't spot or name an em dash if it punched them in the face.
Hi, I’m Jamaal. Welcome to Jamaal’s Letter, where I share insights, predictions, and strong opinions on the future of markets, work, business, tech, and more.
The people who read this newsletter are the people who run the world. If you aren’t yet a subscriber, consider joining them by subscribing below. Please share it with your friends and colleagues if you find it helpful:
The internet is now hunting em dashes. Online sleuths have determined that em dashes — the longer-than-hyphen dashes that appear mid-sentence and often function like commas, colons, or parentheses — are a sure sign of AI writing.
I've long been a fan of the em dash. Since my days acing an editing class at the University of Missouri School of Journalism, to publishing writing in places like TIME.
I write things as if I'm saying them. I'll listen to the spoken version of my texts and make edits when something doesn't "sound right." That often results in a tendency to overuse commas, and the em dash allows me to switch it up every so often, adding variety to my punctuation.
And while I don't let AI write anything for me completely, the truth is that there's little writing I produce that is completely untouched by AI:
"Edit this for spelling, usage, and grammar."
"Give me 10 options for the opening sentence. Here are a few potentials:"
"Add some transitions between sections."
"What are the gaps in my logic?"
"Edit this to be clearer and understandable."
Until its recent fame, most people couldn't spot or name an em dash if it punched them in the face. Decades of texting, messaging, and social media have led to an explosion of informal writing and a decline in punctuation.
Em dashes are ubiquitous in AI writing, because they're ubiquitous in the formal, edited English writing — books, research papers, news articles — that AI has trained on. Em dashes are less a sign of AI than they are a sign of formality, punctuation, and editing — each increasingly rare in today's social media-saturated world.
But I'm now scared to use them, worried that I'll be falsely — or maybe accurately — accused of using AI. I've gone so far as to instruct my favorite LLMs to eliminate em dashes from their vocabulary. I haven't embedded this edict across all of my prompts but I often add it to individual ones.
The irony is that whether you're hunting em dashes or running text through AI detection software, "if you're trying to guess if AI wrote this, you're wasting your time."
"AI has officially passed the Turing test. OpenAI's GPT-4.5 model has done what has long felt inevitable - passed the Turing test in a study by UC San Diego researchers.
In a three-party interaction, participants engaged in 5-minute text conversations with both AI and humans and tried to identify which was which.
The results?
GPT-4.5 was mistaken for a human 73% of the time, surpassing the 50% threshold of random chance.
Among many things, this means you can no longer tell if AI created the:
→ LinkedIn post you're reading now
→ video you're sharing
→ code you're reviewing
→ paper you're grading
→ investment memo you're reading
The software you paid money for to tell you if something is AI?
Garbage.
Every day, I talk to executives, board members, educators, and government officials who are still trying to be the AI police.
If you've allowed your process or organization to fundamentally hinge on knowing AI's role in the creation of something important, you're already behind.
The strategic question isn't "How do we detect AI?" but rather "How do we design systems that remain effective regardless of who—or what—created the input?"
If you need outputs that are objectively beyond AI's reach, you need inputs that AI can't produce."
I think the em dash hunting will be short-lived. The shaming of AI writing is a symptom of AI's "awkward adolescent phase" in which AI is being used rampantly and transparently while being equally distasteful.
In other words, everyone uses it yet many people find it inauthentic.
"AI is in its awkward adolescent phase — simultaneously more transparent and less tolerated than it will ever be again.
Look at how organizations are responding:
↳ If an organization hasn't yet launched AI products or rebranded existing ones, it has at least responded to its board's request to develop an AI plan.
↳ CTOs have earmarked budget allocations for AI initiatives, but these efforts are still largely experimental.
↳ Those "internal" memos discussing AI prioritization and hiring slowdowns (e.g, Shopify) continue circulating, often reading more like press releases than genuine internal communications.
Despite widespread adoption, people still don't like it:
↳ Employers don't want AI job applications. Applicants don't want AI application reviewers.
↳ People on social media are hunting "em dashes" and shaming AI-written content.
↳ One way to be surely rejected in any highly selective process is to be perceived as transparently using AI. And our interaction with AI today is still remarkably explicit:
↳ We type directly into prompt boxes and specifically reference AI in our conversations.
↳ Only a few weeks ago, the entire over-30 internet turned themselves into action figures.
In other words, AI is still novel, shiny, and labeled.
Soon, AI will become more deeply integrated into existing technologies and processes:
↳ We'll transition from typing detailed prompts to simply voicing our demands.
↳ Finding any final content completely untouched by AI assistance will be rare.
↳ Rather than functioning as a separate tool requiring deliberate engagement, AI will seamlessly embed itself into all the tools we already use." This adolescent phase won't last for long.