The Em Dash Dilemma: AI Detection and Writing Style in the Age of ChatGPT

The dead giveaway that ChatGPT wrote your content - and how to work around it

If you use ChatGPT to write emails, essays, and more, you will want to make this small tweak. I have been testing ChatGPT since it launched in November 2022 and have seen updates improve its writing capabilities. However, fairly recently, I noticed a pattern in ChatGPT's output -- the repeated use of the em dash. Also: ChatGPT is your personal shopper now - here's how it works While the em dash is a punctuation mark loved by many writers, such as myself, as seen in the sentence above, it also seems to be ChatGPT's favorite. While the em dash, also referred to as a hyphen, is a sophisticated way to add a pause, break, or an aside into a sentence, it may be worth switching to the comma, or you may start raising suspicions that you rely on AI. Don't believe me? Once I started noticing this pattern, I looked it up, and it's not just me noticing it. There are Reddit threads, OpenAI Developer Community discussions, and even various articles dedicated to what is referred to as the "ChatGPT Hyphen". As I stated earlier, em dash aficionados are everywhere, wedded to their favorite punctuation mark, so I ran an experiment to test my observations. Also: 5 reasons I turn to ChatGPT every day - from faster research to replacing Siri I asked ChatGPT to write me a quick conversational summary explaining what ChatGPT is and to define some use cases. In the first sentence, the em dash made an appearance: ChatGPT is basically an AI that you can talk to -- kind of like having a super helpful writing buddy on call 24/7. It's trained to understand all kinds of questions and prompts, and it can generate text that sounds pretty natural and human. You can use it for tons of writing tasks: things like drafting emails, writing blog posts, coming up with social media captions, brainstorming ideas, or even helping edit and polish what you've already written. It's also great for creative projects like stories or poems, and it can explain tricky concepts in a really easy-to-understand way. Basically, if it involves words, ChatGPT can probably help! I then asked ChatGPT to tell me about New York in the summer, and the em dash was sprinkled through its output: Absolutely! New York City in the summer is vibrant, energetic, and packed with things to do -- but also hot, busy, and sometimes a bit overwhelming. What to expect: It's a mix of grit and magic -- chaotic but unforgettable. Are you thinking of visiting or just curious? Also: 3 clever ChatGPT tricks that prove it's still the AI to beat As I continued my testing, many outputs did not include em dashes, especially shorter answers. However, generally speaking, em dashes are not commonly used in everyday language, and, as a result, ChatGPT's constant gravitation towards them has made it almost a signature of its text. Luckily, there is a super useful trick to keeping your text em dash-free. You can ask ChatGPT at the end of your initial prompt not to use em dashes, or, after the AI generates its response, you can tell it to remove em dashes. Since ChatGPT is generally pretty good at following instructions, the changes will happen immediately once you ask. And if you happen to like using em dashes in your writing, do so at your own risk.

No, an Em Dash Can't Help You Detect AI Text

Summary Em dashes aren't exclusive to AI; some expert human writers also use them extensively. The use of em dashes indicates skilled writing, not AI-generated content. Detecting AI text isn't paramount; quality writing should be valued over spotting AI-generated content. With more and more AI-generated content on the internet, it makes sense that many people want a reliable way to know if a given text is AI-generated or not. Sadly, there is no way to conclusively determine if a text is AI-generated, which leads people to making false connections, or developing the unwarranted belief that they can tell when something is written by AI. Now, my beloved em dash has been caught in the crossfire. What's an Em Dash (And Why Do We Use Them?) The em dash -- the bread of this sandwich -- is a punctuation mark that's sometimes used as a more dramatic comma or parenthesis. It's used to call attention to parts of a sentence -- like this -- and guide the rhythm and flow of writing. It's one of the most versatile punctuation marks, but you'll rarely find it in day-to-day or casual writing. The em dash mostly lives in published books and other professional literature. Some People Think Only AI Text Generators Use Em-dashes Because text generators like ChatGPT have been trained on a huge amount of good writing, you'll see quite a lot of em dash use in the text. Personally, I don't see it as much in ChatGPT text as some people claim, but it's certainly present to a greater extent than the typical person's writing would contain. The thing is, em dash use is something you see from professional writers, which means that believing that usage of the em dash is a sign of AI would lead to plenty of legitimate human writers being accused of using AI to create their content. As one freelancer writer laments in a post on X, the so-called "ChatGPT hyphen" would make her work as a professional writer look like AI to people who believe it's a sign of AI writing. I already worry that professional writers will start avoiding the em dash in their writing, which will almost certainly remove a lot of the personality and craft from their work. It's like asking someone to remove a tool from their toolbox, leading to a situation where you have to use a screwdriver as a hammer. Except, Human Writers Love Em Dashes Too The use of em dashes isn't really something that's emphasized in school, so it's mostly college-educated folks, particularly those who studied language, literature, and the humanities in general, that are most likely to use em dashes in their writing. Here at How-To Geek, we actively encourage our writers to use -- but not overuse -- the em dash. We want writing that has some verve. It has to be entertaining and fun to read -- the em dash can do that and more! Every person who writes for a living can probably admit to getting a little too enthusiastic about the em dash, but it would be a real tragedy if it went extinct simply because of a spurious association with AI, brought on by people who perhaps don't read the sort of literature that typically features this wonderful bit of punctuation. Related You Don't Need ChatGPT Plus to Use Custom GPTs, Just to Make Them My custom GPT doesn't care if I'm premium or peasant. Posts The Tell-tale Signs of AI Text Are More Subtle Insofar as anyone can tell whether something is written by AI, the actual signs that a text comes from a bot might be much more subtle than the presence of a (relatively) common punctuation mark. As someone who edits hundreds of articles every year by writers looking to break into the business, combined with reading a ton of AI-generated content, there are some things that set off alarm bells in my head. Of course, no one can prove that a given text is AI-generated, but in practice it doesn't matter whether something sounds bot-like because a real human writes that way, or it actually does come from an LLM -- we can do better. Of course, a lot of the stuff I personally notice in ChatGPT text, for example, may change or go away as the models advance, but here are some observations: Overuse of the "It's not just X, it's Y." formulation. E.g. "Metallica isn't just a band, it's a musical revelation." A tendency to pad out text with redundant sentences that repeat information we already have, or don't actually say anything of substance. Of course, this is also exactly what inexperienced writers chasing a word count have always done. A weird, impersonal, and inappropriate professional tone. Though some style guides demand exactly this from people too. What makes this task even more impossible, is that you can use clever prompting to make chatbots like ChatGPT drop these bad habits and sound more human. You can even give it samples of your own writing and have it copy your style. Related How to "Humanize" ChatGPT Text One more time, with feeling! Posts Don't Bother Trying to Detect AI Text Ultimately, whether the text you're reading was generated by AI or not is almost beside the point. If the writing you see doesn't read well and is riddled with inaccuracies, it should be disregarded. By simply engaging critically with a text, you'll end up only paying attention to good writing, and if you're of the opinion that AI software can't produce good writing, then it's better than simply picking on em dashes as a way to filter likely AI text out. Related How AI Content Detection Works (and Why It Often Doesn't) AI content may be getting too smart to detect. Posts