The em dash has become one of the most recognizable fingerprints of AI writing, a punctuation mark that instantly makes readers wonder whether a human or a machine composed the text. In this article, we’ll explore how this tiny horizontal line came to dominate modern prose, why AI models rely on it so heavily, and why many readers have learned to treat it as a sign of artificial authorship. You’ll learn how the em dash evolved from its historical roots in typesetting, how it influences readability and rhythm, and why language models like ChatGPT keep inserting it even after being told not to. Finally, we’ll look at practical ways to communicate style preferences clearly so AI tools can produce text that truly matches a writer’s voice.
What is an em dash?
An em dash (—) is a long horizontal punctuation mark often used to create a pause stronger than a comma but lighter than a period. It gets its name from its length, which historically equalled the width of a capital “M” in traditional printing. The em dash emerged in the 18th and 19th centuries when typesetters needed a flexible way to insert dramatic pauses or shifts in tone without breaking a sentence apart.
Writers quickly embraced the em dash because of its versatility. It can replace commas, parentheses, or colons, making prose feel conversational and fluid. In fiction and journalism alike, it became a stylistic choice for signalling emphasis, interruption, or sudden contrast. For instance, the difference between “He stopped, surprised” and “He stopped—surprised” is subtle but real: the latter feels more immediate, almost like a thought happening mid-sentence.
Today, the em dash appears everywhere from novels and essays to online content. But as we’ll see in the next section, that very popularity has created an unexpected problem in the age of AI: readers have begun to associate the em dash not with style or nuance, but with machine-generated text.
Why the em dash has become the signature of AI writing
In recent years, the em dash has developed a strange new reputation online. Once considered a mark of style and rhythm, it is now one of the clearest visual cues that a piece of text might have been written by artificial intelligence. Readers, editors, and even detection tools often treat excessive em dashes as a telltale sign of machine-generated language.
This association didn’t happen by accident. Large language models are trained on billions of words drawn from books, websites, and academic papers, many of which use the em dash liberally. The mark appears so frequently in polished, professional writing that AI systems have learned to treat it as a natural choice for connecting ideas, signalling tone changes, or emphasising contrast. The result is an algorithmic bias: models overestimate how often real people use em dashes in everyday writing.
Humans, by contrast, tend to rely more on commas or short sentences to create the same flow. That’s why, when someone sees several em dashes packed into a paragraph, it feels off — too smooth, too consistent, almost unnaturally balanced. This subconscious signal has trained many readers to spot AI-written text at a glance, even before running it through detection tools.
Why AI (ChatGPT) keeps using em dashes even after being told to stop
When a user asks an AI model to stop using em dashes, it sounds like a simple instruction. Yet, the request conflicts with how large language models are built and how they interpret style. Understanding this behaviour requires a quick look inside the architecture of modern AI writing systems.
Language models such as ChatGPT are trained to predict the next word or symbol in a sequence based on probability. During training, they absorb billions of examples from books, websites, and articles, where em dashes appear frequently in well-edited text. Over time, the model “learns” that inserting an em dash is statistically more likely to make a sentence flow smoothly or appear stylistically refined. Even when users say “don’t use em dashes,” the underlying probabilities remain the same, and the model’s instincts lean toward inserting them because they appear so often in the training data.
Another reason involves tokenization. To a model, an em dash isn’t just punctuation; it’s a token that signals a certain sentence rhythm or semantic pause. When a user instruction conflicts with a strong statistical pattern, the pattern often wins unless the request is repeatedly reinforced within the same context.
Finally, formatting tools and editor environments sometimes convert two hyphens into a single em dash automatically. So even if the model outputs “--,” the final rendered version may show an em dash. This can create the illusion that the AI ignored the user’s request when in fact the issue lies in the post-processing layer.
The combination of these factors makes the em dash one of the most persistent quirks in AI-generated writing.
How to stop AI (ChatGPT) from using em dashes
Telling an AI to “stop using em dashes” often isn’t enough, because the instruction competes with statistical habits learned from millions of high-quality examples. To override that bias, you need to phrase the request in a way that is both explicit and reinforced by context.
1. Set a house style early.
State your rule clearly at the beginning of the conversation or document, such as:
“Use commas, periods, or colons instead of em dashes or en dashes in all cases.”
This sets a firm constraint that the model can apply throughout the session. The earlier the instruction appears, the more influence it has on the model’s output.
2. Replace, don’t just forbid.
If you only say “don’t use em dashes,” the AI knows what not to do, but not what to do instead. By specifying acceptable alternatives, you provide positive guidance. For example:
“If you need a pause, use a comma. If you need emphasis, use a colon or a short sentence.”
3. Reiterate your preference mid-task.
When the AI outputs text containing em dashes, correct it directly rather than vaguely. For example:
“Replace all em dashes with commas or periods.”
This reinforces the rule and resets the model’s context window for the rest of the writing session.
4. Check for auto-formatting.
Some editors, including blogging tools and document processors, automatically convert double hyphens (--) into em dashes. To prevent this, use plain-text editors or disable smart punctuation.
5. Keep style rules consistent.
AI models respond best to consistency. If your request changes between drafts (“no em dashes” one moment and “use more varied punctuation” the next), the model may revert to its default patterns.
By setting clear boundaries, providing alternatives, and maintaining consistency, users can train AI writing tools to follow their preferred style far more reliably.
Conclusion
The em dash may seem like a small detail in writing, yet it has become one of the clearest symbols of how humans and AI approach language differently. To a person, punctuation is a choice of rhythm and expression. To an AI, it’s a probability pattern shaped by data. That difference explains why the em dash appears so often in machine-generated text and why it stands out as a clue to readers trained to spot artificial tone.
Ultimately, the goal isn’t to eliminate the em dash but to understand what its overuse represents: a lack of stylistic alignment between human preference and machine learning. By communicating rules clearly and reinforcing them throughout a session, writers can guide AI tools to produce cleaner, more natural prose that truly matches their voice. In the end, mastering how to talk to AI about punctuation is less about fighting the machine and more about teaching it to listen.
Interested in how AI shapes modern writing? Check out our other articles on ChatGPT, AI tips and tricks, and digital content trends.
Recommended Products
