resident fire truck enthusiast

Obligatory statement that I am not an expert, and all of my claims are based on what I (and friends) have observed over the past year, along with our prior knowledge of what AI-generated text often looks like. I'm working with the material that I have in front of me.

AI-generated summaries are the easiest to notice and the most formulaic. Spotting AI in the summaries also saves you the time of starting to read it, getting like halfway through, and going "...wait a second."

These are examples of what I've come across the most:

“When John [exposition], [inciting incident].” “When John first met Henry, he felt the spark between them immediately. Henry didn’t, which only made John’s feelings more complicated.”

“Between [x], [y], and [z], John must [take action].” “Between coffee shops, nightclubs, and doctors’ appointments, John and Henry are forced to confront the one thing they could never truly escape: Each other.” (^ This one makes me see red. I would say that this is the most obvious sign that a summary is AI-gen.)

“But what happens when [conflict]?” “But what happens when Henry knocks on the door he, himself, had slammed shut years ago?”

“Will John [action a], or will he [action b]?" “Will John open the door, or will he keep it locked as he promised himself he would?”

“What will happen when [x]?" “What will happen when John opens the door?”

"John [action a], Henry [action b], and together, [action c]." "John fell first, Henry fell harder, but neither of them knows what to do with that information."

[Ending the summary in a question that implies more is going on] - "... Or do they?" - "... Right?" - "... But what if it isn't?"

So when put all together, the summary may look like this: “When John first met Henry, he felt the spark between them immediately. Henry didn’t, which only made John’s feelings more complicated. Between coffee shops, nightclubs, and doctors’ appointments, John and Henry are forced to confront the one thing they could never truly escape: Each other. But what happens when Henry knocks on the door he, himself, had slammed shut years ago? Will John open it, or will he keep the door locked as he promised himself he would? Or John fell first, Henry fell harder, but neither of them knows what to do with that information... Or do they?" Does this look familiar to you? Is this ringing any bells? Because half of the summaries in the tag that inspired this post either look like this or contain multiple of these elements, and it's really disappointing.

I think the best way to describe these summaries is the blurb for a Hallmark movie.

Obviously, not everything that's AI-generated is going to look like this (and not everything that looks like this is AI), but I think that this is a good starting point for recognizing what an AI-gen summary would look like, how they tend to be formatted, and the kinds of patterns they often follow. AI-generated summaries of fics on AO3 look like a Hallmark movie blurb because the Hallmark movie blurb is the template.

I can see an argument being made where someone asks “so people can’t use tropes in their summaries without being accused of using AI?” And to that, I say that it’s fine to use tropes, but it’s the fact that all of these summaries I have seen carry the same tone, and the specific elements they have follow these templates almost to a T that makes it incredibly suspicious and worthy of being looked into. One summary like is it whatever. Every other fic having multiple of these elements? Well.

Then, for the "writing" itself, these things I would say are the easiest to spot and are the most common ones I've seen:

“Not [x], not [y], but [z].” or "It was [x]. [y]. [z]." - “The look on John’s face was not angry, not offended, but pained.” - “It wasn’t an accident, not a coincidence, but fate.” - “The barista was not happy, not energetic, but exhausted.” or - "It was comfortable. Calm. Serene." - "He was quiet. Isolated. Alone." - "The house was lively. Warm. Safe."

"[Adjective], [synonym]." - “There were dark, bruised shadows underneath his eyes.” - “He pulled out a faded, heavyweight hoodie.” - “His face carried a gray, translucent quality.” - “The screen lit up, casting a harsh, artificial white glow across his face.” - “There was a silent, mutual agreement.” - "His chest moved in short, guarded hitches." - “The empty space felt vast, cold, and entirely too still.” - “There was a freezing, bitter distance between them.” - “He hadn’t just been tired, he was exhausted.”

As you might have noticed in all of these examples, there is a heavy emphasis on adjectives and synonyms being used to paint a specific picture, which, to be fair, is what normal writing also aims to do as well, but gen-AI often falls back on the same handful of templates for its visualization, and it gets really uncanny when you find it easier to pick up on its patterns.

Fun fact! All of the examples for "[adjective], [synonym]" were pulled directly from a fic I had been skimming through, and of those nine, three of them came from the same three-sentence paragraph. Like, it gets to a point where it starts to feel like a parody of itself.

I feel the need to reiterate that reading something that includes these does not automatically make it AI. Using synonyms to emphasize an adjective is so common in AI-generated writing because of how often it's used in all types of writing (I've used it multiple times throughout this post). The same goes for the rule of three, especially when the third is made to subvert expectations. However, it's the number of times these show up in a short period, combined with a lack of variation among how these ideas are presented, that should raise some flags. You see these once or twice a chapter? That makes sense; they're common devices that are used so often that it's instinctive for a lot of people. You see these devices enough times in a chapter that you start to go "oh, you again?" Something fishy might be going on.

I'm not gonna bring up the em dash/excessive use of commas, since I feel like everyone mentions those in their "keep an eye out!" posts, but also because generative AI has tainted the em dash, and that makes me really mad. I love the em dash. Look what they took from us—now I can't use it without worrying someone is going to call me out on it. Same with commas. You can pry the excessive use of commas from my cold, dead hands.

There are a lot of other tells and indicators that you can definitely find other people talking about, but this post is long enough, and my hands hurt.

And one last thing! I've noticed that when people are told that something they've read and enjoyed was actually AI-generated, they more often than not get really defensive about it and try to deny any of the evidence they are presented with, and though it's frustrating to not have someone listen to you, you also have to acknowledge that it's likely that the person is just feeling embarrassed that they enjoyed something AI-generated and don't want to accept that they did. It's like when you realize you fell for an AI-gen video or picture. You get a little bit ashamed and feel like an idiot. However, I have also seen people who, as a result of their defensiveness, start to harass people who call out gen-AI, which is incredibly shitty behaviour, and shame on you if you have done that. Let's be normal about things!!

And here's the obligatory "you can't know 100% for certain if something you're reading is AI-generated because AI is stealing from actual writing, so publicly accusing someone of using AI in their writing is always a slippery slope and often counterproductive because if you're correct, yes, this person should be ashamed of using AI to write fanfiction (of all things), but if you're wrong, this person you've accused is going to be harassed for no reason, even if your intentions were in the right place."

Anyway, sorry for the wall of text, I got annoyed and had like an hour of my life to spare.