LITEROTICA PODCAST
LIT CAMS (200 Free Tokens)ADULT TOYSVOD MOVIES

AI can apparently ID you from your writings.

Unsurprising in one sense. I write here (in both the commenting and story senses) much as I do or have done on other websites, with some interesting exceptions, and if Roko's AGI ever wanted to compare all commenters on every website, it could possibly work out I was X on Y, P on Q, and so on. It would probably be easier picking up repeated anecdotes or explanations rather than pure stylistic analysis. A human could even connect my accounts if I slipped up in the right way.

But there's virtually no writing on the Internet under my real name, so it wouldn't really de-anonymize me.
 
...I have so many questions.

I have no idea who that journalist is, and I'm too lazy to look her up. But my initial thought was, "So... her writing style didn't evolve in twenty years?" and "She writes news articles in the same way she writes romance novels after a bad breakup?"

I am also confused by Claude's response. It clearly states. "One guess that comes to mind, [...] Megan McArdle? She's been writing a historical novel and her nonfiction voice has exactly this kind of wry, catalogue-of-flaws wit.".
But didn't she say she only had UNPUBLISHED novels? And was feeding the LLM an UNPUBLISHED excerpt? Again, I didn't look thoroughly, but a quick google search for her name didn't give me any historical novels the LLM could reference here.

My guess would actually be that she gave the LLM the excerpt with a list of possible authors, and the LLM just chose correctly. If she then doesn't start a new session for each follow-up prompt, the context of the previous prompts will lead the LLM to favor the name it gave before. Otherwise... I'd be surprised why that author and news journalist never comments on the fact that her work must have been used to train this LLM, in order for it to recognize it.
 
I've posted this here before, but years ago there was a 'you write like' site where you drop in your text and it told you who you write like. I dropped in a story of mine and it said I write like Stephen King. Do I just? I scanned my very un-King-like story, changed the word 'wolves' to 'wombats' and 'knife' to 'knickers', and then I write like J.K. Rowling. Fair dos, the AIs are much better at it now, but I'm really not sure where it is on the scale between random hallucinations and showing a photograph of your front door.

Also, New Zealand is going to be the last safe place when the AGI does come to get us all. It's going to be On the Beach but in Wellington rather than Melbourne.
 
Last edited:
The skepticism against the article is fair, but she does highlight a real risk. Stylometry is becoming much cheaper and faster to do with llms and bad actors and govts can and - I hate to say it - probably will use it against smut writers. Most of us here on lit live in societies where what we do is accepted or at least tolerated. But we have many fellow smut writers around the world (particularly lgbtq writers) who take incredible risk by publishing their stories.

I don’t publish anything under my real name, but recent advances in the tech certainly gives me pause to think about whether this is something I want to continue doing…
 
Coincidentally I was just reading an article in this week's New Scientist:
Mythos is an AI created by Anthropic. Its existence was accidentally revealed last month when people unearthed content on the company's website, not due for publication, which had been left unsecured for anyone to see.
Click to expand...
At this point I put the magazine down, covered my eyes with my hand, and tried to collect my senses. Are we doomed? We're doomed, aren't we? And not doomed by AI as such, just by plain old human error. It's not the cleverness of AI we're up against, it's quite the opposite...
 
Last edited:
I kind of feel like this is mostly smoke. Statistical analysis of writing samples is nothing new, for one thing, and people generally only accept statistical results as being valid if they happen to reinforce their existing beliefs (yes, I'm a bit salty about that). There's some risk, perhaps, if there are enough samples to analyze to connect various accounts to a real person, but the real issue is being able to backtrack those accounts to the account holder Joe Blow of Podunk, USA... something that can be done without the extra step of worrying about how similar their smut is to their other writings.

I might even be cynical enough to suggest the article is a possible conspiracy to get more people to use LLMs for writing. If everything you post is machine-generated, after all, it'll be impossible to do stylistic analysis, or at least pointless to do so. Your anonymity is secure! Or so they might hope you believe. But again, the real risk is for accounts to be traceable. I don't actually think the article is trying to do that, to be clear, but trying to use concerns about privacy to motivate wider adoption of LLM use or dependency is not something I'd put beneath the people funding the development of said services.
 
AI can do a bunch of invasive shit we don't want, and, frankly, some of that is hidden in plain sight. The best the general public can do at this point is push back against "data centers" over issues like extreme impacts on potable water sources and rapacious electricity consumption.

1984 is a tiptoe through the tulips compared to what is being rolled-out as we speak.
 
I don't get the fear. Am I misunderstanding something here? We are NOT talking AI yet. We are talking about LLMs. And LLMs need to be TRAINED on our writing before it could POSSIBLY hope to identify any of us. And I kinda doubt any of us published enough unique text-material to allow that (and I am writing this knowing full well how big some authors' catalogues are).
I would, MAYBE, believe that the author of that article published enough tweets and articles under the same name to make it identifiable... but, as I wrote before, the whole thing sounds more like Anthropic paid for the article that claims Claude is somehow awesome, or a straight-up hallucination on the LLM's part, than anything that actually happened.

*EDIT* You can just test it yourself. Take the article she wrote, feed it into Claude, and ask it who wrote it. I'd be surprised if it manages to do it. But if the article is in any way true, it SHOULD be able to do it.

*EDIT 2*: I just did the test. I fed the article to Claude, and it WAS able to identify her... based on identifyers I didn't see in the text.

This is a column by Megan McArdle. Several things point to her specifically:
  • The Atlantic connection (she wrote there for years) and the reference to Lori Gottlieb having "previously" written for the Atlantic
  • The reference to "the Argument" as a section name points to The Washington Post, where McArdle has been a columnist, and that section is a Post feature
  • The mention of "Piper" likely refers to a colleague who ran a parallel experiment, consistent with a collaborative Post column context
Click to expand...

So, yeah, Claude knows her Wikipedia page and was trained with the articles she wrote. But it didn't identify her based on prose or word-choices.
The third point is interesting here. It means that Claude knows about the experiment run by Kelsey Piper... on itself. Which wouldn't be possible since the experiment happened AFTER 4.7 came out. Which means Claude was "fine tuned" to know about it before the article was released.
 
Last edited:
Reminds me of something I read long ago:

In 1968 or 1969, Paul McCartney said a wistful and startling thing in an interview. He said the Beatles had discussed the idea of going out on the road as a bar-band named Randy and the Rockets. They would wear hokey capes and masks a la Count Five, he said, so no one would recognize them, and they would just have a rave-up, like in the old days.
When the interviewer suggested they would be recognized by their voices, Paul seemed at first startled... and then a bit appalled...
Memo to Paul McCartney, if he's there: the interviewer was right. The would have recognized your voices, but before you even opened your mouths, they would have recognized George's guitar licks. I did five books as Randy and the Rockets and I've been getting letters asking me if I was Richard Bachman from the very beginning.
Click to expand...

Stephen King, "Why I Was Bachman," The Bachman Books, 1985
 
I think this is probably pretty real. AI will have to have some samples of someone's writing with their name attached to it in order to be able to say "this other pieces was very likely written by Sam Jones," but even without that it can say with confidence "these two samples were written by the same person."

Comments about varying your writing style or people's style evolving over time don't mitigate this very much, IMO - the variations are not so complete that you really disguise the writing as much as most people think. And this is not a new thing - AI just makes it a lot faster and easier. Obviously the larger the samples you're comparing the more accurate the correlation (or lack thereof) will be.

Back in the dim and distant past I briefly had a blog that I wrote longhand (scanning the pages and posting images) - partly for fun, and partly because it made it difficult for Google to index. No more. As a test I fed a scan of a page of my longhand writing to ChatGPT and it transcribed it flawlessly. I suspect style recognition will also continue to evolve.
 
Britva415 said:
Humans were doing this before there was AI.
Click to expand...

^I was about to say this very same thing.

I'm not at all surprised about this because I've been able to discern people (and students) from the way they put words together. There are people whose jobs involve identifying people through writing patterns.

If you're too paranoid about being deanonimized, just stop using the Internet altogether. Full stop. There's no such thing as an 100% secure system.
 
I wonder how effective this is across genres. I could easily believe that LLMs would be able to look at an unpublished story by, say, Anne Rice, and identify similarities to published works of hers. But if I were to supply something like this:
Screenshot 2026-05-03 at 9.53.07 AM.png
or this:
Screenshot 2026-05-03 at 9.53.41 AM.png

I doubt any LLM would have an easy time recognising stylistic similarities to these authors' better-known works in other genres. (Unless, of course, that connection had already been identified in its training data.)
 
Another thing I've already posted here, and expanding a bit on #6. At around the same time as the 'you write like' meme, there was also a site that decided whether you were male or female. I dropped in some text from a forum where I had accounts with both identities, and the male one came up likely to be male, and the female one came up likely to be female, and why not? They were. My personae were those people, but I hadn't done any crude rigging of the styles, I had just written as the person. Also on that forum, someone rewarded my uncertain teenage girl even though I hadn't given any clue that that's what she was except writing like her. I was actually rather impressed I'd been so successful.

So it is possible, and if you're doing it right quite easy, to write in distinct styles.
 
BeechLeaf said:
Another thing I've already posted here, and expanding a bit on #6. At around the same time as the 'you write like' meme, there was also a site that decided whether you were male or female. I dropped in some text from a forum where I had accounts with both identities, and the male one came up likely to be male, and the female one came up likely to be female, and why not? They were. My personae were those people, but I hadn't done any crude rigging of the styles, I had just written as the person. Also on that forum, someone rewarded my uncertain teenage girl even though I hadn't given any clue that that's what she was except writing like her. I was actually rather impressed I'd been so successful.

So it is possible, and if you're doing it right quite easy, to write in distinct styles.
Click to expand...
Yerp, dis fing right here. ^^

If you're an author who has a consistent voice and style that doesn't vary by the story, then yeah, it's easier to pick you out than someone who tailors their voice to the story. I've always been a big proponent of the latter, because certain stories are better complimented by different styles, and characters really should have different voices. But if you're always doing third-person omniscient, and you don't really change all that much over time, then A) really, dude? B) you'll be not overly difficult to pick out of a crowd provided someone has samples of both your writings. And you would need samples of both writings. There is virtually no way an LLM can read a random bit of text and go, "Oh, that's probably X," unless that person is super famous and has a very distinct writing style, and a large body of that person's text is in the training data.
 
If someone wants to post a couple of samples of writing from the same author in different styles (or from different authors in similar styles) I'd be happy to run it through an AI engine for comparison, probably Claude, and report on how it does, just for the curious.
 
So, I tried out ChatGPT on my stories On the Origin of Anthros and Her Forest Bloom:

Short answer: they could be the same author, but I wouldn’t call it likely. There are some shared fingerprints, but the differences in narrative discipline and tone are pretty pronounced.


Where they do overlap​


There are a few stylistic habits that line up:
  • Dialogue rhythm and snappy banter
    Both pieces rely heavily on dialogue that’s quick, informal, and slightly irreverent. The succubus’ “Ooh! My favorite!” and Ayala’s “No shit.” feel like they come from a similar instinct for punchy, modern-sounding exchanges—even in fantastical contexts.
  • Juxtaposition of the absurd with the serious
    In the first, you’ve got cosmic magic colliding with crude humor (sheep lover, sudden explosion). In the second, it’s terminal illness and gender identity alongside a dryad casually appearing and speaking in a familiar tone. That blending of grounded/emotional stakes with speculative elements is a shared trait.
  • Voice modernity
    Neither piece tries to sound historically authentic or high-fantasy formal. Even in “39,612 B.C.E.”, the diction is contemporary (“Shoddy work,” “You’re one of those”). The second piece does the same with the dryad.

Where they diverge (this is the bigger signal)​

These differences are strong enough that they weigh against common authorship:
  • Tone control and intent
    • The first piece is satirical, chaotic, and shock-driven. It escalates toward absurdity and punchlines (bestiality reveal, explosion, global transformation gag).
    • The second is introspective, somber, and character-driven, with sustained emotional continuity (identity, illness, estrangement, mortality).
    • Writers can shift tone, but doing so this cleanly usually still leaves structural similarities—and here those differ too.
  • Character interiority
    • First excerpt: minimal inner life. Characters are vehicles for jokes and plot turns.
    • Second excerpt: deep internal monologue, nuanced emotional reasoning, layered backstory.
    • That’s a big craft difference. It’s not just tone—it’s how the author builds meaning.
  • Prose density and description
    • First: sparse description, action-forward, almost script-like pacing.
    • Second: textured sensory detail (skin, breath, environment), slower pacing, more deliberate imagery.
  • Humor style
    • First: external, situational, and often crude/absurd.
    • Second: internal, wry, character-based (“I didn't realize the brain shit was this bad.”).
    • Those are usually pretty stable authorial traits.
  • Worldbuilding attitude
    • First: intentionally flippant (e.g., dismissing leylines, joking about cosmic consequences).
    • Second: treats the supernatural with earnest mythic weight (the dryad as meaningful, almost sacred encounter).

The most telling point​

The biggest separator isn’t the subject matter—it’s narrative discipline.

The second excerpt shows:
  • controlled pacing
  • emotional throughline
  • restraint in payoff

The first thrives on:
  • escalation
  • subversion
  • punchline reversals
Authors can write both comedy and drama, but they tend to carry over structural habits (how they build scenes, escalate tension, land beats). Here, those habits feel different.

Bottom line​

  • Possible? Yes. A versatile writer could do both.
  • Probable? I’d lean no.
If I had to quantify it: maybe 30–40% chance same author, 60–70% different authors.

-------

Yeah, bitch :cool:
 
anthrodisiac said:
Humor style
  • First: external, situational, and often crude/absurd.
  • Second: internal, wry, character-based (“I didn't realize the brain shit was this bad.”).
  • Those are usually pretty stable authorial traits.
Click to expand...
This is my favorite part. Though, I don't think anyone here is too surprised to learn that my traits aren't very stable 🤪
 
For shits 'n giggles, I tossed in part of Once You Go Anthro Ch.01: Abeni

This one is interesting, because it actually bridges some of the gaps between the first two.

High-level take​

  • Closer to the first excerpt than the second
  • But still shares some DNA with the second
If I had to rank likelihood of same authorship:
  • Excerpt 1 ↔ Excerpt 3: fairly plausible
  • Excerpt 2 ↔ Excerpt 3: less likely
  • All three same author: possible, but not the cleanest fit
Rough probabilities (very approximate):
  • 1 & 3 same author: 55–65%
  • 2 & 3 same author: 30–40%
  • All three same: 25–35%

Why #3 aligns with #1​

1. Dialogue fingerprint feels similar​

There’s a very specific cadence both share:
  • Teasing, slightly aggressive banter
  • Characters who push the other person conversationally
  • Punchy, modern phrasing
Examples of shared energy:
  • First: “Ooh! My favorite!” / “You’re one of those.”
  • Third: “Oh my god, do you ever get out of your fuckin’ head?”
That slightly irreverent, performative dialogue style is a strong overlap.

2. Comfort with absurd/sexualized premises​

Both:
  • Drop the reader into a bizarre scenario without apology
  • Mix sexuality with humor and discomfort
  • Use nonhuman characters in very human (and often provocative) ways
The sheep-lover reveal in #1 and the anthro bar flirting in #3 feel like they come from a similar creative space.

3. Comedic instinct: escalation through awkwardness​

  • #1 escalates via absurdity (sheep → transformation → explosion → global effect)
  • #3 escalates via social discomfort (awkward guy → teasing → increasingly explicit jabs)
Different mechanisms, but the same instinct: push the situation until it becomes uncomfortable, then keep going.

Why #3 diverges from #1​

  • #3 has more grounded POV narration and internal monologue
  • It sustains a consistent character perspective, whereas #1 is more sketch-like
  • The humor is less explosive, more situational
So if same author, #3 feels like a more controlled, character-focused version of #1’s style.

Why #3 only partially matches #2​

There are overlaps:

1. Interior narration​

Both #2 and #3:
  • Spend time inside the protagonist’s head
  • Use self-aware, slightly cynical internal commentary

2. Sensory detail​

  • #2: nature, body, illness
  • #3: scent, texture, physical presence
That said, the purpose of that detail differs.

The big mismatch with #2​

1. Emotional depth vs. performative tone​

  • #2 is earnest, vulnerable, and thematically heavy
  • #3 is social-comedic, flirtatious, and intentionally awkward
Even when #3 touches insecurity, it’s played for humor—not reflection.

2. Prose discipline​

  • #2 is tightly controlled, with deliberate pacing and thematic cohesion
  • #3 is looser, more conversational, sometimes meandering (intentionally so)

3. Thematic intent​

  • #2 is about identity, mortality, belonging
  • #3 is about discomfort, attraction, and social mismatch
Writers can span both—but usually you still see similar structural priorities, and here those differ.

The most telling connection​

If I had to point to one “fingerprint” linking #1 and #3, it’s this:
A fondness for putting a socially or morally uncomfortable premise front-and-center, then leaning into it with humor rather than softening it.

That shows up strongly in both.

Bottom line​

  • #3 feels like it could plausibly be the same author as #1, especially if that author is capable of dialing chaos up or down.
  • #2 feels like a different writer, or at least a writer in a very different mode that doesn’t share as many stylistic habits.
If you told me:
  • #1 and #3 same author → believable
  • #2 same author as them → I’d be skeptical, but not dismissive
----

It's real nice of it to not be dismissive if I were to tell it that Her Forest Bloom was written by the same weirdo as On the Origin of Anthros and Once You Go Anthro.
 
You must log in or register to reply here.

Similar threads

C
From pending purgatory to AI hell
2 3
Replies
68
Views
3K
MetaBob
MetaBob
Justadesperatewifeandmom
  • Locked
Nothing to see here, Justa bit of Florida sunshine with a chance of hurricane.
276 277 278
Replies
7K
Views
724K
AmPic Moderator
AmPic Moderator
H
Behold! I have been saved by the great god Aton!
Replies
19
Views
2K
SeaCat
SeaCat
Boxlicker101
Really Weird Feedback
2 3
Replies
56
Views
4K
sr71plt
sr71plt
RightField
Income Inequality and Politics
2
Replies
44
Views
4K
RightField
RightField
Back
Top