Notes from the Loft on Em-Dash Panic
On AI Detection and the Fear of Sounding Human
I started listening to Sam Harris’s interview with Susan Cain, someone I admire enormously as a fellow disciple of quiet living. One of the early topics of discussion between them is the messy, but very prevalent topic of AI writing. Cain tells Harris that sometimes she finds herself engrossed in an article online only to discover along the way one of ‘the tells.’ As you may imagine, this was very interesting to Harris. He wanted to know more: What are ‘the tells’? That is a question that many people are asking. Common replies include the em-dash and the so-called triplet. I became immediately skeptical as she named these ‘tells.’
When people speak about ‘tells’ my brain immediately goes to: oh no, they’re fishing. As someone who is highly neurodivergent, I’m already insecure about everything I do. I try not to compare myself to others, but it’s extremely hard to avoid it. Now we have to worry about writing that is too perfect? Does perfection, or something near it mean it must have been produced by a machine? I thought AI writing was bad. That’s what people used to say. I’ll never forget the librarian who was quoted in American Libraries Magazine in the summer of 2025. She was talking about AI-written books found in the children’s section of the library: “The writing is laughably bad,” she said. Laughably bad! A year ago, that’s what was being said. Now the writing is too perfect?
History is full of moments where intuition hardened into accusation. Moral panics thrive on shifting criteria.
But what about the Em Dash tell? I use the Em Dash a lot. I like it. It breaks up a sentence. It breaks a train of thought. It makes a turn. I’m sure that’s why Virginia Woolf liked it. I’m currently reading Jacob’s Room, and let me tell you, it is full of em dashes! In Woolf, the em dash is not decoration. It marks interruption, hesitation, association, drift. It is part of how thought moves across the page. It signals the flow of consciousness itself. Maybe that’s why it has become an early scapegoat of AI panic.
Underneath the skepticism lies a deeper fear: that the LLM will learn to hesitate like us, to mimic not just performance but faltering.
AI learned to write by reading us. Why are we surprised when it resembles us?
I know it has to be frustrating. Uncertainty rules the world. Nothing is stable. AI comes along and destabilizes it further. That’s unsettling. What’s going to happen to human writing? Will the humanities survive? How will it impact authors and readers and libraries? Are we doomed? What can we do to save humanity?
I worry that our panic (maybe even our best intentions) will drive us to do more harm than good. The moral panics of the past were dangerous. They destroyed reputations. They upended lives. There is the potential for a book that is written by a human to be misidentified as AI-written and, consequently, targeted for a ban. Imagine making a mistake like that? Banning a book you think is written by a machine because you spotted a ‘tell’? Are you ready to assume that kind of responsibility?
I don’t know what the answer is. I’m only certain about uncertainty. These ‘tells’ people are talking about feel dangerously unreliable. They aren’t scientific or quantifiable. They are heuristic, intuitive, and highly subjective. (That sentence is a triplet, isn’t it? That might be AI slithering its way into my article? Who can say?!)
I struggle every day against my own errors of judgement. I endeavor not to write self-consciously but authentically and freely. I know we all have to face criticism. As writers, we have to learn to live with it, to use it, and learn from it. But I don’t see the benefit of writing in an atmosphere of social pressure to be more or less than what I am just because it might be mistaken for AI modality. If my humanness has to become performative, I’m not sure I would know how to do it. I’ve never been a good performer. I’m afraid I would do it very badly.