How to find out if a story is generated by an Ai or written by someone

[Zenomew]

Just a random thought that hit me

How can you tell if a story or piece of writing is AI-generated or written by a human? Are there specific signs or patterns to look out for? Also how do you spot if a story was written by an AI or just a writer who ran out of steam and decided to let ChatGPT do the heavy lifting while he filled out the rest

 

[Assurbanipal_II]

They have a certain style and wording. And they lack a human touch.

 

[Zenomew]

They have a certain style and wording. And they lack a human touch.

What wording and style exactly

 

[SailusGebel]

I know AI mixes up pronouns a lot. Other than that, I have no idea.

 

[Assurbanipal_II]

What wording and style exactly

I can't say how, but the writing feels monotone ... Uninspired. And Robotic. And the wording is imprecise. In sense, the AI avoids repetitions like hell.

"What wording and style exactly"

An AI for example would never forget the question mark.

 

[Astrolust]

I think a lot more authors use it than we know. With Sudowrite or it's alternatives, I'm not sure you'd even catch someone doing it if they took the time to properly edit it. I low key could care less about the AI aspect as long as the story is engaging, almost no plotholes, interesting characters, etc. The real problem i have is someone will just get on chat gpt with no concept of prose, and just be like "Computer generate Trump and Chris Christie Yaoi fanfic" And then just keep having gpt generate more and more content. The ai on its own loses track of itself. but if you put a proper person in charge of keeping track of the story i bet it would read better than 3/4's of the slop korean smut MTL Yandere story's i read.

 

[Jocelyn_Uasal]

Websites like this exist, though just like how every author has their own unique voice, AI writing has its own voice too that just so happens to be almost exactly the same from piece to piece.

 

[Tempokai]

AI (or rather GPT based LLMs) love to rush things. They overgeneralize, try to make a distant narration style, have overly optimistic endings (even if it's bleak story), hate gore, hate body horror, have specific opening and closing lines. Generic, over the counter (lol) LLMs have zero personality on purpose and sound like the most dry math school teacher.

ChatGPT is simply information without any frills. Specifically tuned ones can bypass the "AI feelness meter" (which I just invented from my brain), but it takes a lot of attempts and proper tuning to do so. NovelAI's is simply put, random. Inconsistent, has lots of pecularities, but mostly will have the style the text is written on. If the text veers off into randomness or randomly repeats the previous text, you found one. There's a new version of it, though, maybe it improved somehow.

The "AI detector" stuff just parces the most probable tokens, and puts them in whatever "humanity scale" they've had come up with, which i'd say a coin toss levels of accuracy.

I'd say you can find if the text is AI generated by reading a lot of AI written stuff (like I did) and based on "feeling" of it find if it's AI or not. It's already a future, but nobody's ready for it.

 

[Ruti]

How can you tell if a story or piece of writing is AI-generated or written by a human? Are there specific signs or patterns to look out for? Also how do you spot if a story was written by an AI or just a writer who ran out of steam and decided to let ChatGPT do the heavy lifting while he filled out the rest

for the most part you can just... tell
AI don't tend to be that good at stories, and if you read enough it becomes obvious which is or isn't AI. Though that's only applicable for the lazy ones who put AI then leave it. There is occasionally the one who uses AI, then edits what the AI types, so it becomes harder then. Eitherway, AI is AI, and AI tends to do the same things because of how their trained.

 

[Bluebery]

if you find these phrases, the story is AI assisted:

- characters name Elara
- barely above a whisper
- otherworldly glow. (AI lovessss to give everything glowing in the light when describe things)
- a testament of

there probably many other phrases but these are just the most common ones

 

[beast_regards]

There are AI detectors, but those are unreliable. They sometimes even flag works written before invention of the AI ...

AI isn't that difficult to spot in writing if you understand what it really does.

First, AI isn't really Artificial Intelligence, not even remotely. It doesn't know what it is doing. It is merely a sophisticated script which calculates the probability that the one letter is followed by another, and spouts the words, without being aware of the content or context. "The AI" always loses the track of its own writing because it doesn't know what the words mean, and only lays them, without even the trace of intelligence, and in time, it is apparent after the while. It usually does something the human poor writer wouldn't. The continuity errors are very common and often feel jarring because they could appear in the frequency the poor (but still human) writer wouldn't do. Grammar checkers (which are "the AI" practically) also don't understand concepts and continuity, but they perform considerably better as someone with human brain is doing it.

A Google Translators, however, would always feel like AI. If your first language isn't English (my isn't) and you feed this through the AI translator, you would always get something that gets it off track. The "Korean novel" argument is wrong, imho, it's just a mechanical translator doing stupid thing, even if the writer is human. Spellcheckers are sometimes guilty of the same thing.

Second, the AI is a prude, like Royal Road mod on the ritual bath water. They will never do something edgy, sexy, or whatever. They will never, never do this "Computer generate Trump and Chris Christie Yaoi fanfic" and always refuse, or misunderstand.

All AI content is heavily curated, and when I say heavily, I mean HEAVILY. People try to feed it weird stuff all the time, and corporations literally have hundreds of the minimum wage slaves to comb through it.

So as a rule of thumb: AI writing is always curated content with no context and no continuity.

AI visual art (i.e. pictures) are in fact considerably better in fooling you, and it is common that the artists are accused of using AI even if it was their hand drawing all along, but they too lack consistency and context. They are, however, sometimes considerably less curated as they often run on the separated, downloaded stuff.

 

[RecursiveDescent]

AI also tends to use big words in places where it's unnecessary.
Does describing a simple scene really need to sound like a science paper giving excruciating detail instead of keeping it simple and emotional?

 

[CharlesEBrown]

AI visual art (i.e. pictures) are in fact considerably better in fooling you, and it is common that the artists are accused of using AI even if it was their hand drawing all along, but they too lack consistency and context. They are, however, sometimes considerably less curated as they often run on the separated, downloaded stuff.

Something I'm seeing increasingly are human initiated works - written or graphic - that are then run through filters and wind up looking or feeling almost organic but with a strange "sameness" that is more disconcerting than anything else.

 

[beast_regards]

- otherworldly glow. (AI lovessss to give everything glowing in the light when describe things)
- a testament of

I had used "otherworldly glow" in my story. I think I even used "testament of..."

You know, I am angry with you now. You may have disturbed my insidious plan to take over the entertainment industry, but just wait ... I am redirecting my mind control satellites!

 

[LoneQuack]

if you find these phrases, the story is AI assisted:

- characters name Elara
- barely above a whisper
- otherworldly glow. (AI lovessss to give everything glowing in the light when describe things)
- a testament of

there probably many other phrases but these are just the most common ones

Two out of the four you listed, I've used myself... I believe (hopefully) you are not implying I'm an Ai but just because something is uncommon doesn't mean its Ai (referring to 'barely above a whisper' and 'otherwordly glow' although the otherwordly not pared with glow)



For the moment, you can tell if you analyze it enough. For the moment... Give it two more years, and we can have the convo again.

 

[beast_regards]

Something I'm seeing increasingly are human initiated works - written or graphic - that are then run through filters and wind up looking or feeling almost organic but with a strange "sameness" that is more disconcerting than anything else.

I don't think it's a case for the visual art.

This whole AI "scandal" started with the journalists running the images through filters to harp on how AI steals art. Though it certainly happens, in an attempt to avoid the common errors the AI did (and still does), the AI could produce unique looking (even if ugly) designs back then. Now I believe it is because the databases it uses are more established and widespread, and many people use the same LoRA resulting in sameness of the picture as well. I don't think most AI generated pictures are run through filters these days, just using the handful of same resources. It wasn't the case back then. In fact, if you use the AI image generator online (I did, because I can't set up it) it still does the usual from the early days, and there are only a few ways to get guaranteed results.

The writing, however, it is definitely the case. The spell checkers do have rephrase function, and ProWritingAid and Grammarly are experts at Corpo-talk. So is ChatGPT for what matters.

 

[ConansWitchBaby]

From the one story I have read that uses it, it tends to repeat half a sentence at least twice within a few paragraphs. It also repeats the same information after a few paragraphs too.

You need to see the information hook of the writing and see if it reiterates the same thing or expands/withhelds a continuation of it as a human.

 

[ChiriVulpes]

If you want to be able to catch AI writing, in general you're just looking for excessive amounts of purple prose. You won't spot it in dialogue, it'll be in big blocks of narration. Anything that sounds excessively wistful, repetitive stuff talking about the situation the protagonist's in being a huge deal for them, anything that sounds excessively hopeful for the future or the character's "journey" in the middle of a chapter. Keep in mind that most people using an AI for writing are just feeding prompts into something like ChatGPT and then just lightly editing whatever the output is, so you're very likely to get like 4-5 paragraphs from the AI, where in the last couple the AI is trying to "wrap up" whatever concepts it was given with the prompt.

 

[CharlesEBrown]

I don't think it's a case for the visual art.

This whole AI "scandal" started with the journalists running the images through filters to harp on how AI steals art. Though it certainly happens, in an attempt to avoid the common errors the AI did (and still does), the AI could produce unique looking (even if ugly) designs back then. Now I believe it is because the databases it uses are more established and widespread, and many people use the same LoRA resulting in sameness of the picture as well. I don't think most AI generated pictures are run through filters these days, just using the handful of same resources. It wasn't the case back then. In fact, if you use the AI image generator online (I did, because I can't set up it) it still does the usual from the early days, and there are only a few ways to get guaranteed results.

The writing, however, it is definitely the case. The spell checkers do have rephrase function, and ProWritingAid and Grammarly are experts at Corpo-talk. So is ChatGPT for what matters.

Either I was unclear or you misunderstand me (or, perhaps, both) - what I'm seeing is people starting with real photos or hand written text, and running THAT through AI filters. The original product was very different, but the end result looks suspiciously the same every time.

 

[beast_regards]

So far, I didn't find many novels that abuse the AI to the point it is noticeable to the reader.

Most of us (writers) here on the Scribble Hub, and other similar sites, including even the ones like the Royal Road, are amateurs without any formal education or training in writing, and we do what we think is best to look professional, purple prose included.

I thought I did encounter a novel on the Royal Road which may be AI generated, but only because of unrelated blowback which have nothing to do with the AI, and was simply the sad reality how the RR works for everyone.

I found one novel (or Author, rather) that is obviously AI generated to the point it is felt. Audiobooks are, as far I could tell, professionally narrated, it doesn't seem like the mechanical text to speech, but the content itself seen some AI writing to the point I couldn't unsee it.

Either I was unclear or you misunderstand me (or, perhaps, both) - what I'm seeing is people starting with real photos or hand written text, and running THAT through AI filters. The original product was very different, but the end result looks suspiciously the same every time.

I saw it happening for a text.

The rephrasing via the ProWritingAid does that, Grammarly also tries to force a specific way of speaking since it hates creative writing and thinks I write the corporate quarterly report. It's technically AI, but not quite, since it doesn't create things from prompt, but it does change the tone, and everything. Usually I don't want to sound like the higher management after too much cocaine, so I don't use it.

Not sure for the visuals.