You can spot AI-generated content from a mile away. Not because of what it says, but because of how it pauses. That little horizontal line that shows up between clauses, the em dash. ChatGPT loves it. Obsesses over it. Uses it like punctuation confetti.
Why? Because of how it was trained.
The RLHF bias
ChatGPT was fine-tuned using Reinforcement Learning from Human Feedback (RLHF). Human reviewers rated outputs, and the model learned what got high scores.
Turns out, reviewers loved clarity. They rewarded outputs that broke complex ideas into digestible chunks. And one of the easiest ways to chunk a sentence? The em dash.
Instead of writing "The business struggled because of inconsistent branding," ChatGPT writes "The business struggled, inconsistent branding made it hard for customers to trust them."
The em dash becomes a crutch. It signals "here comes an explanation" or "let me clarify that last point." Reviewers saw this as helpful. The model learned to overuse it.
The clarity trap
Em dashes do add clarity. They create natural pauses that help readers parse complex ideas. But overusing them makes writing feel robotic, like the author is constantly stopping to explain themselves.
Real writers vary their sentence structure. They use periods. Commas. Semicolons. Short sentences for impact. Longer ones for rhythm.
AI defaults to the dash because it's safe. It works in most contexts. But safe is boring.
The tell
Here's how you spot it in the wild. If a piece of content has more than two em dashes per 200 words, it's probably AI-generated. If those dashes appear in a predictable pattern (statement, dash, clarification), it's definitely AI.
Examples from actual ChatGPT outputs:
"Local SEO is changing, AI overviews are replacing traditional search results."
"The best brake shops understand their customers, they respond to reviews and post regular updates."
"Your website needs structure, schema markup helps AI engines understand your content."
See the pattern? Every sentence could be split into two, but instead, the dash wedges in an explanation.
How to break the habit
If you're using AI to draft content (and you should, it's a powerful tool), here's how to scrub the dash obsession:
1. Split sentences
When you see an em dash, ask if it should be two sentences instead. Most of the time, the answer is yes.
Before: "We optimize for local search, GBP and schema markup are critical."
After: "We optimize for local search. GBP and schema markup are critical."
2. Use conjunctions
Instead of dashes, use "because," "but," "so," or "and." These create flow without the robotic pause.
Before: "Your reviews matter, they're the first thing customers see."
After: "Your reviews matter because they're the first thing customers see."
3. Prompt for variation
When you draft with AI, explicitly tell it "No em dashes. Use varied sentence structure." This forces the model to reach for other punctuation tools.
4. Edit by reading aloud
Read the draft out loud. If you're pausing awkwardly at every dash, rewrite. Your voice will catch what your eyes miss.
Why it matters for your business
If your website, blog, or marketing copy reads like ChatGPT wrote it, you're signaling to customers (and Google) that you didn't invest time in your own messaging.
Google's E-E-A-T framework prioritizes content that shows experience. AI patterns scream "no human edited this." That kills trust.
Your competitors are using AI too. The ones who win are the ones who make it sound human. That means editing the dashes, varying the rhythm, and adding your actual voice.
The bottom line
AI is a drafting tool, not a publishing tool. If you're posting AI outputs directly to your site without editing, customers notice. Google notices. The em dash is just the most obvious tell.
Use AI to get the bones of your content. Then edit like a human. Break the dash habit. Your brand voice will thank you.