Sorry about the em dash. Really.
AI is beginning to lose some of its shine in the public eye. Some brands have begun labeling their social media and ad content as AI-free to avoid accusations and backlash. Indeed, a 2024 study found that readers preferred content labeled as “human-generated” by 30% over content labeled as created by AI — regardless of what actually wrote it. Open LinkedIn on any given day, and you’re sure to find users bickering about what they believe is a sure-fire sign of using generative AI to write posts.
Many of these conversations are overly simplistic. LLMs were trained on human writing. Anything AI spits out was done first by a human. In many cases, the AI tells are just hallmarks of good writing, just repeated so often and in such a robotic way that things like the em dash move from simple punctuation mark to something sinister. And that same 2024 study found that people aren’t very good at telling actual AI-generated content apart from something written by a person.
But it still matters that communicators are aware of how the use of certain punctuation, phrases and sentence structures can raise doubt about humanity in an audience’s mind, which can ultimately harm trust over time as people begin hunting for evidence of AI rather than absorbing your messages.
This is not to say that everyone should stop using the word “delve” or never employ a negative parallel structure. It’s simply to say that everyone should be aware of the perception of certain writing styles so they can employ them intelligently and build in other hallmarks of authenticity. And if you are using generative AI, maybe edit out some of these hallmarks.
Here are a few quirks of AI-generated writing to be on the lookout for — in your own writing and beyond:
- Repetition of sentence structure.
Generative AI has a few favorite sentence structures. One sign to look out for is when every sentence reads the same, sentence after sentence, paragraph after paragraph. Humans vary sentence length and structure for drama and excitement. AI often uses short declarative sentences. It finds them clear and easy to read. So, it uses them over and over. Until everything sounds the same.
To avoid this trap, simply mix up your sentences. Keep some short, sure. But it’s all right to let others sprawl and dig into a more complex idea over a few joined clauses. Or wouldn’t you enjoy throwing in a rhetorical question, just for fun?
- Repetition of certain
AI’s love of “delve,” a perfectly fine word, has become a meme. In fact, a number of words favored by AI were shown to have seen a sharp uptick in human speech in podcasts and videos just after the widespread release of ChatGPT, showing that these tools are actively changing how we speak. According to Scientific American, these words were:
- Delve
- Boast
- Inquiry
- Meticulous
All of these words are just fine to use. They’re good words, even! But you may want to limit your usage to avoid accusations of AI use.
- Negative sentence structure.
Ask AI to generate a paragraph of text for you right now. I will bet you a shiny penny that it includes at least one use of the structure: “It’s not X, it’s Y,” or “this isn’t just X. It’s also Y.” These negative parallel constructions can be a jazzy one-two rhetorical punch. But AI tends to beat them into the ground until everything is defined by what it’s not. New York Times Magazine writer Sam Kriss said the pattern can drive him “to the point of fury.”
This is a structure to be especially careful and thoughtful when using today. Make sure you’re using it sparingly for emphasis. Once per piece, max.
- The infamous em dash.
Everyone knows AI favors the em dash, but no one is quite sure why. Here’s an interesting lineup of possibilities, ranging from token efficiency in computing to the preferences of the humans doing the quality checks on models. One interesting theory is that many LLMs are trained on older books, written in the late 19th and early 20th century, which use a lot more em dashes than we do today, which makes it stand out much more when machines start peppering in the punctuation.
Ultimately, it doesn’t matter why the em dash is beloved by AI — it matters that it’s become the poster child for AI-generated text. It’s a shame for a really lovely punctuation mark that can always add a dose of drama to any sentence.
The lesson comes back to the same thing we keep addressing: be thoughtful and judicious in your use of the mark. Make them count. They shouldn’t be showing up in every paragraph.
Don’t let AI stop you
AI shouldn’t “ruin” any part of English for any communicator. But it is a vital reminder to slow down, vary our sentence length, structure and word choice. Because those are ultimately the best fundamentals of writing anyway.
Allison Carter is editorial director of PR Daily and Ragan.com. Follow her on LinkedIn.
The post 4 reasons your writing accidentally sounds AI-generated (and how to fix it) appeared first on PR Daily.
