• We've issued a clarification on our policy on AI-generated work.
  • Our mod selection process has completed. Please welcome our new moderators.
  • The regular administrative staff are taking a vacation, and in the meantime, Biigoh is taking over. See here for more information.
  • A notice about Rule 3 regarding sites hosting pirated/unauthorized content has been made. Please see here for details.
  • Due to issues with external spam filters, QQ is currently unable to send any mail to Microsoft E-mail addresses. This includes any account at live.com, hotmail.com or msn.com. Signing up to the forum with one of these addresses will result in your verification E-mail never arriving. For best results, please use a different E-mail provider for your QQ address.
  • For prospective new members, a word of warning: don't use common names like Dennis, Simon, or Kenny if you decide to create an account. Spammers have used them all before you and gotten those names flagged in the anti-spam databases. Your account registration will be rejected because of it.
  • Since it has happened MULTIPLE times now, I want to be very clear about this. You do not get to abandon an account and create a new one. You do not get to pass an account to someone else and create a new one. If you do so anyway, you will be banned for creating sockpuppets.
  • Due to the actions of particularly persistent spammers and trolls, we will be banning disposable email addresses from today onward.
  • The rules regarding NSFW links have been updated. See here for details.

People are using AI but not disclosing it. How do you detect it ?

Probably dialogue when it's just super bland and too on the nose all the time.
 
Want to add that in stories set in established fictional worlds Ai will just Make Shit Up with no relation to canon yet beyond simply flavour.

Much of it reads like those overly alt-verse fanfictions where the author really needs to go write their own original work rather than cramming every single cool idea they've ever had into a single fanfic.

An example of this (without pointing fingers) would be something like Ai adding an entirely new system of magic to a Harry Potter fic in an overly convoluted, flowery description that eventually has no impact on the plot and is never seen or used again.
(But the pseudo magi-babble 'sounds' really cool.)

Eg: you have a HP character "enchant" an item and Ai suddenly needs the blood of three wolves, a potion of vitality and special made enchanting desk to do it when HP wizards quite literally just wave their wands.
 
For a very specific category of giveaway, Claude 4.6 likes to talk ambiguously but pretentiously about characters' emotions in an indirect way, but without any of the other elements that a good writer would use for complicated feelings, like physical symptoms or confused actions. It's hard to describe as a pattern, but it's pretty obvious once you've seen it a few times.

I suspect it was at some point reinforcement trained away from obvious "tell instead of show" writing about feelings, but nobody bothered to train it on actual good writing patterns to use instead, so it instead just gets ambiguous and goes in circles writing a fancy version of "Character X was having Big Feelings, and here is more and more text about how big those feelings are but also no actual details about them so that it seems deeper because you have to guess at it".
 
Last edited:
I find it easy to tell because they all read the same. Because the writing is built around clever phrasing instead of concrete observation. It stacks abstract metaphors, every sentence feels like a punchline, and wraps thoughts up a little too neatly. They read like they are trying to sound insightful rather than just describing what's happening because AI patterns "interesting language" rather then a person's thoughts.

Other signs are all characters having the same "voice" and every conversation feels like they are they are in "serious conversation mode" at all times.

Another thing is Triadic Structures. You see all the time in AI writing. "it wasn't strength, it wasn't might, it was inevitability" or "he wasn't the same man, he was sharper, colder, someone else" AI uses them too frequently, makes them too symetrical and packs them with abstract or generic words.

As someone who uses AI to clean up my punctuation and help smooth out some of my ideas, it definitely helps a dabbler like me. Some people are using it as a story generator and that's when you can really see the slop. For example here's something I wrote that was cleaned up by AI:
"She bites her lip in concentration, her expression somewhere between curiosity and awe. "It's not healing in the usual sense — it's correction. His body is actively rewriting itself toward a perfect state, and it's doing it step by step. I'm watching it in real time. Strange… He doesn't have a Corona Pollentia. I don't know how but, he's not a parahuman. Whatever it is, it's natural. Or close to it.""
Vs.
She bites her lip, her expression a delicate balance of curiosity, fascination, and quiet awe. "This isn't healing. Not truly. Not completely. It's something else. Something deeper."

Her gaze sharpens, intensity building as she watches. "His body isn't repairing itself, it's correcting itself. Refining. Perfecting. Every cell, every function, every flaw… rewritten. Adjusted. Aligned. Step by step, moment by moment, toward an ideal that shouldn't exist."

She exhales softly, almost reverently. "And I can see it. All of it. In real time."

A pause. A shift.

"Strange. There's no Corona Pollentia. No sign, no signal, no trace. He isn't a parahuman. He shouldn't be capable of this. He shouldn't even be possible."

Her voice drops, quieter now, edged with something uncertain. "Whatever this is, it isn't artificial. It isn't external. It's intrinsic. Fundamental. Something woven into him at the deepest level."

Another pause, slower this time.

"Natural… or something pretending to be."

See what I mean?
 
I find it easy to tell because they all read the same. Because the writing is built around clever phrasing instead of concrete observation. It stacks abstract metaphors, every sentence feels like a punchline, and wraps thoughts up a little too neatly. They read like they are trying to sound insightful rather than just describing what's happening because AI patterns "interesting language" rather then a person's thoughts.

Other signs are all characters having the same "voice" and every conversation feels like they are they are in "serious conversation mode" at all times.

Another thing is Triadic Structures. You see all the time in AI writing. "it wasn't strength, it wasn't might, it was inevitability" or "he wasn't the same man, he was sharper, colder, someone else" AI uses them too frequently, makes them too symetrical and packs them with abstract or generic words.

As someone who uses AI to clean up my punctuation and help smooth out some of my ideas, it definitely helps a dabbler like me. Some people are using it as a story generator and that's when you can really see the slop. For example here's something I wrote that was cleaned up by AI:
"She bites her lip in concentration, her expression somewhere between curiosity and awe. "It's not healing in the usual sense — it's correction. His body is actively rewriting itself toward a perfect state, and it's doing it step by step. I'm watching it in real time. Strange… He doesn't have a Corona Pollentia. I don't know how but, he's not a parahuman. Whatever it is, it's natural. Or close to it.""
Vs.
She bites her lip, her expression a delicate balance of curiosity, fascination, and quiet awe. "This isn't healing. Not truly. Not completely. It's something else. Something deeper."

Her gaze sharpens, intensity building as she watches. "His body isn't repairing itself, it's correcting itself. Refining. Perfecting. Every cell, every function, every flaw… rewritten. Adjusted. Aligned. Step by step, moment by moment, toward an ideal that shouldn't exist."

She exhales softly, almost reverently. "And I can see it. All of it. In real time."

A pause. A shift.

"Strange. There's no Corona Pollentia. No sign, no signal, no trace. He isn't a parahuman. He shouldn't be capable of this. He shouldn't even be possible."

Her voice drops, quieter now, edged with something uncertain. "Whatever this is, it isn't artificial. It isn't external. It's intrinsic. Fundamental. Something woven into him at the deepest level."

Another pause, slower this time.

"Natural… or something pretending to be."

See what I mean?
That's the thing, wether it has been generated by AI or 'cleaned up' by AI it's still so noticeable. Feels like dozen other ai fics across all platforms and people are sick of it. I myself wrote some AI content, like I wrote stuff and AI cleaned it up for me and after a lot of time where I felt like I couldn't be proud of the work... I refuse to say it's mine since it isn't I stopped. All that to day that generated or cleaned up doesn't make a difference, it's still the same kind of slop
 
The best heuristic I've found is that they're going to use generated AI images for their characters eventually. If they do, you can assume they'll have no qualms about the rest either.
 
@Irdiumstern Not really a good comparison. Some can't draw and do not have the fonds needed to commission an artist. It's still AI sure but in the context of fanfics and writing it's largely different.
 
@Irdiumstern Not really a good comparison. Some can't draw and do not have the fonds needed to commission an artist. It's still AI sure but in the context of fanfics and writing it's largely different.
Writing doesn't need images. You can do what fanfic authors have done for decades now and look for fanart or celeb pics if you think it necessary.
It's a moral hazard either way, and someone justifying one way will flip flop on the other soon enough.
 
Writing doesn't need images. You can do what fanfic authors have done for decades now and look for fanart or celeb pics if you think it necessary.
It's a moral hazard either way, and someone justifying one way will flip flop on the other soon enough.

Taking it kinda far you know? Like saying just because someone smokes they will definitely do drugs and commit crimes.

Just because one thing is true doesn't mean the other will too.
 
The way I see it, if I can still detect it, it means the writing isn't good enough, whether it's AI or not. If I can't detect it, even if it's AI-assisted or whatever, then the writer did a good job.

I know several fics that use AI. I still read one of them because the story is interesting enough and it's a rare crossover. But the others don't have that luxury.

As someone who've used AI to generate wacky crossover stories, I know exactly the kind of writing style it defaults to if you prompt it without further style instructions.
 
So. Many people in a lot of website are using AI to "write" stories these days. QQ has not been spared. I can feel that some stories use AI but I have no proof. And I don't want to throw shade at someone and it not being AI.

Th worst thing is that the premise of some of these stories are interesting but when you read it you get a flash bang of AI slop.

I am not counting in people who just use AI for proof reading.

How do you prove that a story is AI ? I personally just look at the tempo and punctuation. Also the usage of em dashes is a sure sign but not a definitive one. AI stories have a certain measured way of writing. It takes a lot of time to move from one scene to another.

I am tired of investing myself in a story and then start suspecting it's AI. Will there ever be a solution ? I am begging for the AI bubble to pop.
Heard that AI has "Hallucinations" where it makes up a certain amount of content it generates. Not sure about the validity about it, but its something.
 
This is an extremely complex topic because there is no 100% reliable way to distinguish AI generated text from original text written by a human. There are very specific markers that can be identified besides, well, very cheap generators that do not support the sequence of events and the text.

1. Triple mention. This has already been discussed and it is in https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing , but AI usually limits itself to three points when describing something.
2. Comparison through negation. This is currently the most natural and adequate way to spot AI text, since all major generative models like to compare in the style of "The warrior was not just tall, he was a mountain."
3. Natural division of sentences into separate groups and excessive use of commas. This is somewhat harder for me because English is not my native language and I naturally tend to use more commas, but there is also reverse logic here as AI divides sentences into more adequate and shorter blocks instead of writing monstrous constructions several lines long as some authors do.
4. "Serious speech." Literally "This is effective" as a phrase and the antithesis "This is dishonorable" or "This is wrong." Even informal characters will not say "If it works - it works" or similar phrases. AI strives to make the text more "literary" and this is noticeable.
5. Either physical or psychological traits of characters get lost in battle scenes. Without prompting it is difficult for AI to hold and mix dialogues, moral tension and narration. Usually either the personal description or the characters' actions in the process get lost.
6. Hard lines at the end of generation. This appeared not so long ago and only with top models, but they have been taught to finish the text within the task block. Like a miniature cliffhanger or some one-liner joke, you will clearly see it as a piece of several hundred words that starts with some joke or detail and then this detail repeats roughly after 1200 words.
7. Absence of typos and errors. We are all human and text on the Internet is usually not of ideal quality. You will see missing letters that do not fully affect the meaning, lost periods and such small things that even beta reading misses. I personally like to look medical equipment advertising brochures for work and I have a score of about 50/0 wins that every advertising brochure has at least one error or typo. AI never makes "simple" errors, but sometimes gets confused in complex grammar or punctuation.

Oh yes, even without points one can note the "flavor" of top AI. At least I can more or less guess GPT, Gemini, Sonnet/Claude. Each of them has its own work specifics and some stable speech patterns or speech style that catch the eye. However, this works only if the author does not know what Jailbreak is, since strong JBs not only cancel corporate bullshit but can also significantly "humanize" the text. Characters start swearing, speaking in short phrases, the text loses literary strictness, etc. In general, even if the author of the AI text is so lazy that he does nothing himself, a strong JB and maintaining a set of instructions and a "glossary" of the work will not let the AI text be spotted so easily.

To our happiness, 90%+ of AI authors do not even understand how to work with AI text generators.

P.S. Recently I read on SV and SB several large Warhammer 40000 fanfics obviously written by AI and no one noticed it, which is strange. I am not sure how it is in other settings, but AI really copes poorly with deep Warhammer lore, since it preserves the visual style but not the details that are obvious to fans and very noticeable when reading. Has anyone noticed this for other settings that have their own complex lore with a bunch of details?
By itself it's not (I also write using those sometimes). But when I can see too many sign I know it's AI edited or created.
This is a separate personal pain of mine because it was incredibly pleasant to use alt+0151 to clearly indicate dialogues. It differs from the regular dash in the text, so it was convenient to simply type with the long dash, and then during editing one press and you see all the dialogues for work.
 

Users who are viewing this thread

Back
Top