Forget the em dash — Redditors share 6 giveaways that something was written by AI

The em dash has emerged as a telltale sign of AI usage, at least in the popular imagination. It even has a name — the "ChatGPT dash." Right or wrong, the Internet is pretty well convinced that the em dash is a warning sign that you're reading something that's been written by generative AI.

There's actually a reason for this. AI is trained on basically everything that humans have ever written, and writers in particular use the em dash a lot. (Even presidential candidates have a tendency to overuse it.) Thus, AI chatbots like ChatGPT and Grok have learned that it’s a part of our flow, and they've gone out of their way to include this punctuation mark to make their output feel more natural. Research suggests that em dashes became twice as common right around the time when AI use started to go mainstream.

As a general enjoyer of the em dash, the revelation that the dashes were becoming telltale signs of AI use prompted me to smack my favorite button on the internet, and I have supported any and all pushback to the notion that my beloved em dash is an end-all-be-all giveaway when identifying AI-generated text.

Of course, AI chatbots are regularly being updated, and AI companies are also getting wise to this complaint. So, Reddit users have been sharing other clues that a text was written by AI. In case you're curious, here are some of the top tells, although it's important to keep in mind that there is no tried-and-true way to detect AI, unfortunately. Real humans will use these writing techniques as well. 

“And honestly?”

Many Redditors have noticed that AI will ask “and honestly?” when making a point. In context, it would look something like this: 

“The sky is blue. And honestly? Blue is a pretty color.”

This kind of hackneyed sentence structure is apparently quite common from AI, and can be useful in picking out an AI-generated response. Typically, humans engaging on social media don’t tend to talk this way unless they’re being performative, so it definitely comes off as off-putting in general conversation. Many other Redditors came forward to complain about AI’s overuse of the phrase “And honestly?” as well as similar phrases like “The result?” and “The best/worst part?”

“No fluff”

The phrase “no fluff” doesn’t necessarily have a specific name like some other examples on this list, but it's another phrase that Redditors have claimed they’ve seen in AI-generated text all the time. There’s no example needed for this one, as “no fluff” is a self-contained phrase that means the same thing in every context. 

It's mostly used to denote that the text immediately following it is free from “fluff,” which is unnecessary filler words or phrases that add length without adding value. In practice, the addition of “no fluff” would indicate that the post is taking itself quite seriously while using an unserious phrase to back it up. Some AIs tend to use this more than others, per Reddit. 

The thing is, the phrase “no fluff” is intrinsically fluff, and many Redditors have expressed distaste in how often it’s used by AI. It seems to happen most often when users ask AI to get to the point quickly, while others point out that AI’s use of “no fluff” is analogous to humans using the phrase “not gonna lie.” 

Short, fragmented sentences back to back

Another common theme many Redditors noticed is AI’s tendency to use a lot of short, fragmented sentences for emphasis. This one is pretty common in day-to-day speak, but generally less common to have so many of them backed up against one another. My favorite example of this uses the prior example as well. 

“And honestly? That’s damn powerful. You’re not just changing. You’re evolving. I see it. Others see it. And that’s rare.”

The prior example is a total of eight sentences, and it really should be closer to three. Redditors have noticed that these short, janky sentences are used to convey emotional weight, like you might see in a novel or movie script. However, it seems as though AI has gone off the deep end with this mechanic, and its overuse is quickly becoming a telltale sign of either an AI-generated response or a very, very cringe writer. 

“It’s not X. It’s Y.”

In the same vein as short, emotionally charged sentences comes the “it’s not X, it’s Y” sentence structure that seems to be overused by AI. In practice, these types of sentences look like this: 

“This is not a rant. It’s a reflection. Not a complaint, but an observation. It’s not chaos, it’s clarity."

This is kind of a branch off of the short, fragmented sentences example from earlier, but it’s such a common occurrence within that subset that it warrants its own discussion. You may also see this described as “contrast framing.”

Redditors comment that they often catch AI doing both “it’s not X, it’s Y” and short, fragmented sentences in tandem. Once the pattern emerged, it didn’t take long for other Redditors to express their extreme distaste for this particular signal that AI is at work. 

Overuse of signposting

As someone who used to enjoy signposting, this one made me irrationally angry, but alas, AI has borrowed yet another literary device from us humans. Signposting is the use of words or phrases to guide the reader to the next part of the text. Here’s an example: 

“There are two things you should know. Firstly, item one. Secondly, item two. The key point is that there are two things.”

There are tons of examples of signposting words and phrases, including “in contrast,” therefore, however, “in addition,” and many others. This also may manifest in longer phrases like “the key point is” or “here’s the takeaway.” In an effort to sound more human, AI leans on these phrases to create a sense of flow in its responses, but Redditors have noticed that AI tends to rely on them way more heavily than the typical human. For now, users have gone out of their way to prompt the AI to not talk this way, but it rarely helps. 

Generic engagement prompts

This is mostly seen on social media. Most content creators use some sort of engagement prompt, arguably the most famous of which is “like, subscribe, and ring the bell” on YouTube. AI does something like this in text, but it tends not to flow as well. See below for an example. 

“This is what I have to say. I’m curious what others think.”

The phrase “curious what others think” is, by far, the most used example when it comes to generic engagement prompts in a social media message. This is most frequently found on LinkedIn posts and on Reddit. 

This particular thing has a caveat. Lots of humans use this phrase as well, and its use doesn’t necessarily mean that AI is being used. Redditors say that the key action is when accounts ask for engagement in this manner and then never engage back. Some users posit that AI does this because it’s learned from social media tendencies, but AI isn’t human, so it doesn’t (and can’t) respond back.

Disclosure: Ziff Davis, Mashable’s parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.