I'm not a cop.

Purely for statistics, if you are unfamiliar, here is a per-chapter word count list for “One of These Nights”:

Chapter 1 posted on July 18th — 9k words Chapter 2 posted on July 21st — 14k words Chapter 3 posted on July 23rd — 17k words Chapter 4 posted on July 26th — 11k words Chapter 5 posted on July 29th — 18k words Chapter 6 posted on August 1st — 22k words Chapter 7 posted on August 11th — 32k words

From this I conclude that, since posting Chapter 1, within just 24 days, walshpilled has cumulated roughly 116k words. NaNoWriMo’s standard “novel in a month” pace is 50,000 words in 30 days, which would give us roughly 1,667/day. Stephen King has famously targeted 2,000/day.

The average for “One Of These Nights” is approximately 3x the NaNoWriMo pace. Good read on The Guardian journalist Tim Jonze’s NaNoWriMo experience for context, if you yourself have never participated. In his article, Tim Jonze wrote, “It was a revealing experience—the only way to really hit the deadline was to write without ever looking back, without editing, without even reading what had gone before.”

Yes, because then there are the realistic typing/throughput ceilings. Average adults type about 40 WPM; proficient ones can do 65–70+ WPM (i.e., 2,400–4,200 words/hour if you were literally fucking transcribing, monkey-see, monkey-do). Drafting fiction is slower than simply typing something, something, something. Some research has to be done. And then edits commonly run, you know, some varying ballpark amount of time. Anyone who has ever written anything can easily imagine that hitting the aforementioned numbers for “One Of These Nights” while actually delivering coherent, decently formatted chapters is extremely hard for relatively small time intervals.

In the Chapter 2 notes, walshpilled writes, “I got home at 1am last night and proceeded to write this entire chapter in a post-bar haze, so enjoy.”, (complimented by her tweet “got home from the bar at 1am and then just stayed up until 6am just now writing cowboy mowalsh so chp 2 coming tonight or tomorrow im impatient sorry”), which explicitly insinuates that 14k words were written in one single night + some allotted sleep + some presumed average time for editing. 

And then, in the notes of Chapter 3, “I was dying from period cramps for the last two days so I have been holed up in my room cranking this out.”, which explicitly insinuates that 17k words were written within roughly 48 hours, subtracting some hours of sleep and editing as well. 

Speed alone does not prove AI use whatsoever. But what makes this seem so likely AI-assisted to me is the convergence of this pace, unusually uniform, low-variance style and phrasing patterns across tens of thousands of words that I will delve into a bit later. 

On Twitter, August 5th, walshpilled wrote:

“long paragraphs and uploading frequently are indicators of ai generated fanfic !!!” what if i hit u over the head with a wooden plank. anyways punishing people who have pg degrees that teach you how to write a ton quickly or are just able to lock in and crank out 10k+ a day is STRANGE !!!!! and it shows u have no idea how writing abilities are developed in other careers besides the creative space !! last one sorry but the long paragraph thing is so stupid. u would not last one day in a law class if you think long paragraphs or sentences are signs of bad/ai writing. some ppl learn writing techniques from a non-creative space/job and that will show in their fics. dear GOD !!!

No idea who is punishing who. An important note though, which I would like to mention first, is that long paragraphs, “big” words and em-dashes are by no means AI giveaways. This is not a post about “tee-hee em-dashes”.

This analysis considers whether the text exhibits signs of AI-assisted authorship by examining the writing style, diction, and structure for telltale markers often associated with AI-generated prose. I looked for repetitive phrasing, generic or formulaic language, unnatural sentence constructions, overused patterns (e.g. specific transitions and/or punctuation), hedging or redundancy, and consistency or abrupt shifts in tone for known LLM stylistic fingerprints, as these are the most widely used models for writing. 

Of course, to be thorough, I did actually perform a close reading of the entire story. This also included searching the text for repeated phrases and structures (using keywords like “if only”, “the kind that”, “not just”, “like it was”, etc.), counting occurrences of hedging words (e.g. “almost”, “slightly”, “nearly”), identifying any tonal or POV shifts that might suggest stitched-together prompts or prompt excerpts or an inconsistent voice, and comparing observed patterns to known AI-generated text characteristics documented in discussions. Good informative discussion here, if you are personally unaware of these details to look out for. And here for a quicker read, though I highly suspect this article was also a victim to the clankers’ grubby hands.

The story describes characters’ visceral responses in repetitive ways. Both Samira and Emery’s stomachs “flip”, “twist”, or “drop” whenever they feel attraction or shock, and hearts “sink” or pulses “spike” in numerous instances. For example: “Samira rolled her eyes, ignoring the way her stomach flipped…trying to push away the feeling”, and “Emery’s stomach dropped.”. The word “stomach” appears over 30 times in the context of emotional flutters or drops. While romance writing does include such metaphors, the frequency here feels extremely redundant. AI generators are known to repeat these kinds of emotional cues to convey characters’ feelings, rather than varying the imagery. The result is a somewhat generic emotional lexicon.

Some dialogues have an almost mirrored structure. This tit-for-tat phrasing and sentences that echo each other can sometimes result from iterative prompt completions or edits stitching two AI outputs, which then leads to symmetrical banter. While it could be a deliberate stylistic choice, its frequency gives a bit of a contrived impression; as if the dialogue was engineered for effect. Several scenes (the barn loft confrontation, the bar argument) have this too-neat back-and-forth structure that simply feels mechanically constructed.

The recurrence makes the prose feel highly patterned. Generative AI models oftentimes latch onto certain formulations when generating long texts, leading to inadvertent repetition.

Beyond repeated structures, the story’s diction sometimes skews toward generic or “templated” phrasing that lacks a human author’s unique “voice”. The language feels like “anonymous” prose, if that makes sense. It lacks distinctive quirks that will always betray a writer’s personal style. For example:

The narration employs broad and somewhat cliched descriptors: e.g. calling Emery’s smile “slow and smug” or a voice “oozing arrogance and heat”. Phrases like “her voice had no heat behind it”, “bone-tired”, “loose-limbed and content”, or “spine against the worn leather” pepper the prose. They read smoothly, yet feel familiar and unoriginal, and pulled from a common library of fanfiction phrases — and as AO3’s data was recently scraped, they may be. AI models, trained on large swaths of internet fiction, will reproduce and spit out such commonly used phrasings.

Emotionally distant or clinical narration: Despite being a passionate romance, the text at times describes intimate scenes in a detached and orderly way. For instance, during a heated confrontation in Chapter 4, the prose enumerates actions clearly but without much personal flourish: “Samira shoved her, hard, palms flat to Emery’s chest, but Emery barely moved. She just grinned, slow and feral.”. The sequence of short, matter-of-fact sentences here create a fast pace—and also gives a sterile play-by-play feeling. This can be indicative of AI output, which often tells actions in a straightforward way and can miss the deeper emotional inflection an actual human writer might convey. (In contrast, a few paragraphs later the narrative inserts Samira’s internal fury and desire, but the baseline reporting remains evenly detached.)

Presence of “template” lines that AI often uses in fiction: For example, “It was almost peaceful until the rumble of tires over gravel broke it.”. Starting a scene by describing peace then interrupting it is a classic narrative device, but the phrasing “It was almost peaceful until…” is the kind of line a model like ChatGPT can and will produce to transition scenes.

Transitional phrases and asides reminiscent of AI: The story uses transitions like “And yeah, maybe…” or “No, this was something different.” and rhetorical questions in narration (e.g. “What the hell am I doing here?” early on). These are not unnatural by themselves, but the way they’re deployed feels very consistent and even-keeled, lacking the idiosyncrasies or offbeat humor a writer might foresee themselves, and consequentially, inject. For example, “Nights like these made her selfishly wish driving in the city was this beautiful.”. This is a perfectly coherent sentence about the night sky, but reads as a generic reflective aside. Polite, articulate musing. AI often defaults to this, remaining absent of a strong personal voice; well-written but non-specific and a bit impersonal unless explicitly prompted to delve further.

Another noticeable aspect is the heavy use of hedging words and qualifying phrases, which give the sense of the text modulating every description. This is typical of LLM models. They will hedge to avoid overstatement. In the story, characters and narration frequently use words like “almost”, “slightly”, “nearly”, “just”, “kind of”. While people do speak like this, AI writing is known to insert this imperceptibly to avoid absolutes (e.g. “just enough to make her gasp” instead of “enough to make her gasp”).

Hedging can make writing feel less confident. In “One Of These Nights”, the accumulation of hedges and qualifiers gives the sense that the text is talking around sensations and actions instead of stating them plainly. This is the “talking around the point” quality often noted as an AI trait. A human author will occasionally inject such nuance, but here it’s extremely prevalent and therefore non-committal.

Now, onto the em-dash. I love the em-dash and have used it within this post multiple times already, but it has unfortunately become a documented hallmark. 

AI models tend to deploy em-dashes frequently; also unnaturally, because they mimic written text patterns from training data. The interesting bit is that they overdo it. In casual fiction, one might expect a mix of punctuation (commas, semicolons, ellipses for pauses, etc.). More or less of one or the other, depending on the style. This is just a God-given average and unique for everyone.

While em-dashes are common in fiction, the density here should be noted: 460 prolific use instances in the text.

One might expect a long fanfic written over weeks to show some evolution in style or noticeable shifts in tone between chapters (especially given the author’s notes about writing some chapters in different moods or circumstances). However, the tone remains remarkably steady throughout: wry inner monologue, lush physical description, banter. This uniformity is a bit unsettling. An AI model would maintain the same “voice” unless prompted otherwise, and here we see that even as the story transitions from slow-burn and angst to erotic scenes to softer romantic moments, the narrative voice hardly changes.

The story occasionally shifts POVs between Samira and Emery (e.g., a mid-chapter scene in Chapter 4 suddenly follows Emery’s perspective as she stews in jealousy). While the POV change is marked by a scene break, what’s notable is that the writing style doesn’t really change with the POV. Emery’s internal narration uses the same kind of sentences and tone as Samira’s. Both perspectives are written with the same vocabulary and rhythm. In human-authored dual POV romances, often the voices feel at least slightly different to reflect each character’s personality. Here, apart from differences in what the character notices (Samira notices medical details, Emery notices rodeo details), the narrative feels uniform.

From Chapter 1 to Chapter 7, the prose quality and style stay on the same track. There are no abrupt improvements, regressions, or experiments in the writing. Authors sometimes find their footing or change things up (even slightly) over the course of such a long work, especially if written serially. This consistency can be reached via a very attentive editor (though no beta reader/editor has been noted, so I’m going to assume walshpilled does it herself), smoothing everything down to one tone, which indicates meticulous self-editing to avoid any variance at all—which is uncommon. There are no swings in style, not even by an increment, which does reduce the feeling of human touch.

The writing style doesn’t significantly shift during fights, hot encounters or introspections. The pacing of sentences and the way paragraphs are structured remain comparable. In intimate scenes, I, as a reader, might expect a lot more subjective, maybe even fragmented narration or a change in vocabulary to convey heightened passion. Yet, the sex scenes (though explicit in content) are narrated with the same clear, step-by-step descriptive style as the action scenes. They are explicit, yes, but also somewhat methodical rather than each scene organically finding its own rhythm.

The vocabulary isn’t especially esoteric or anything, but it does avoid slang and remains in a neutral-professional register even when insults or banter occur. I saw minimal use of contemporary slang or highly colloquial dialogue; even the smack talk is, like… relatively tame. The metaphors and similes used (dust clinging to air, being a barn cat basking, etc.) feel like they could come from any competent writer, they aren’t tied to the unique in-story experience. AI, which obviously lacks true personal experience, just produces generic imagery. No highly specific or quirky metaphors that a human being might derive from their own life or a distinct creative angle. Everything is just sterile levels of on-brand for a western rodeo lesbian romance.

Now onto the more important matters: here, either ChatGPT’s o3 or 4.5 (recently launched a 5 version) model might have been used, which is currently only accessible via a monthly fee, or the much worse scenario, which would be Claude’s Opus 4 (recently launched 4.5), which is also behind a paywall and much harder to detect, especially if tampered with.

Both Claude Opus 4.5 and OpenAI’s GPT-4.5 are/were capable of producing fluent narrative prose, but they exhibit very different stylistic tendencies. Read more on this here; nice article. Claude is widely noted for having natural-sounding language that reads a lot more like a human and avoids the stiff, overly formal style. Claude offers more voice and nuance, excels even at quirky and sassy dialogue and injecting humor; giving the white boy some flavor, so to speak.

Claude’s outputs are more concise and free of obvious AI clichés or repetitive boilerplate phrases. This expressiveness can translate into more emotionally rich storytelling. By contrast, early GPT models had this tendency to use super dry, academic-sounding language or overly polished “HR-speak” in all contexts. ChatGPT’s responses are generally more templated and too wordy no matter the prompt.

Untrained readers will struggle to differentiate a well-camouflaged AI story from a purely human-written one, especially as these models improve. I understand the empathetic wish to protect writers when they are wrongly accused, which is actually precisely why I wrote this post—I do not think it was an entirely wrong accusation.

This post is not to explicitly say that all of “One Of These Nights” is all AI-generated, but I do believe it is heavily AI-assisted at the very least. Take this post with a grain of salt. What’s in question is the pattern here, because I actually seldom saw any “big” words being used, no paragraphs were “long”, and, well, the em-dash is just gonna keep em-dashing, as it’s a regular punctuation tool. Taken together, the most parsimonious account is material AI assistance. Writing + editing speed alone proves nothing; speed + the aforementioned fingerprints + a live writing-and-editing-and-posting schedule at ~3x NaNo pace and the same scaffolds and hedging loops showing up again and again, at this scale, does. 

You know, walshpilled, if it is AI-assisted, please be honest and just say so; that’s not a moral failure. I know it would be fucking embarrassing as all hell to admit, and you most likely will not. In my eyes, until there’s literal counterevidence, the cleanest hypothesis is some degree of material LLM involvement. The other option, of course, is that you would just try harder at hiding it.

So no—“I use big words because I have a law degree” isn’t actually the point, and it isn’t responsive. Readers can enjoy the story and still ask for honesty about the tools that were used in creating it.

On to the next story.

I looked at only Chapter 1 of her new fanfic, “The Sicilian Defense” and immediately saw the same heavy hedging and patterned paragraphs… Particularly enjoyed “something [verb]ed or something-ed”; this construction appears 23+ times in just Chapter 1. I’ve even been kind enough to pull up some examplezzz! I love pulling up statistics when it’s time to consider AI slop. The output, without fail, is so algorithmical! Anyway,

“something that was not quite nerves”

“Something twisted in her chest”

“Something flickered in her eyes”

“Something cracked”

“Something mischievous twisted in Samira’s chest”

“something caught on something inside her”

“Something about the way...”

“something hollow ache in her chest”

“Something close to amusement”

“something close to intrigue”

“Something flickered in Emery’s gaze”

“something dangerously close to intrigue”

Why are we not ever committing to what the emotion actually is? And once again, “the kind that/of…”  appears 15+ times:

“the kind that made her stomach twist”

“the kind that felt like a slow suffocation”

“the kind that made her head spin”

“the kind of space reserved for donors”

“the kind of soulless space”

“the kind of look”

“the kind of chaos”

To avoid being redundant here (see: my analysis of “One Of These Nights”), I’ll literally just list the phrases, and you can go through them on AO3 if you’d so desire. “As if/as though” appears 31+ times, “before she could…” appears 12+ times, “like it was/like she was” appears 27+ times… I’ll kindly remind you, this is all just about Chapter 1 still.

Now, onto a more personal topic: I want to see the world where a 25-year-old assistant’s and 37-year-old chess champion’s internal narrative structures are completely identical. These women should think, observe, and process the world fundamentally differently; however, their internal voices use the same sentence complexity, the same hedging patterns, the same physical staging vocabulary, and the same metaphor types. There are no distinguishing markers between characters; particularly glaring in a romance where understanding each woman’s distinct perspective should be central to the narrative. No age-appropriate language differences asides from texting manner in the latter chapters, no palpable class/wealth vocabulary distinctions, no profession-specific thought patterns, no distinct rhythms… Writers typically differentiate POV characters through vocabulary choices, sentence length, internal rhythm—factually everything. Because they are vastly different people. This fic shows none of that differentiation and reads to me as the same generic person was just inserted for both women. Logically speaking, if we are striving for hyperrealism with all this worldbuilding, why do these women have no personality?

The fundamental issue, of course, isn’t simply that Samira and Emery sound identical, because I do not want to allow for this to get sweeped under the rug as “teehee i’m just not a good enough writer :3” or any other copout nonsense similar to that. Neither woman feels like a fully realized person. Both characters are collections of surface-level traits and mechanical reactions. I’m sure you’ve got an inkling as to why that is so!

Writers, you know, they spend time genuinely inhabiting a character’s perspective, especially if they are the focal figures of the story, and make sure that character develops organic complexity. AI cannot genuinely explore these depths. It can simulate the surface markers of characterization, but it has no access to embodied reality. It produces characters who perform actions and experience generic emotional responses, but lack the first step, the skeleton, and that only comes from a writer truly asking: Who is this person? How does her specific life experience impact how she sees this moment? What does she notice that no one else would? And so on, and forth, and around, and through.

The result is characters who exist as flat narrative functions. Samira needs to be “talented but overlooked” so the plot can happen, and Emery needs to be “dominant but intrigued” so the romance can happen. Beyond these functional requirements, both women are empty vessels being moved through scenes while experiencing the same rotation of throat-tightening, heat-crawling, stomach-dropping reactions.

This is why the romance feels soulless despite the explicit content. We’re watching two 100% interchangeable templates collide. Even more so, a lesbian writer genuinely exploring these characters, who are women, would give them texture, specificity, contradiction… The characters lack depth because they weren’t genuinely written and explored. They were generated through algorithmic pattern-matching that can simulate competent prose but does not have the key, which is called “using your brain”, that could access the deeper works.

And that’s a topic I could delve deeper into, but anyway. Back to the algorithm, baby. The text is, once again, same as in “One Of These Nights”, full of perception filters that create distance between the… whatever experience is had, without actually naturally inhabiting a character’s perspective. The consistently “told” narrative; indeed, the density here is to be noted! “She could [verb]” holds 52+ instances, “she seemed to”, 18+ instances, “she tried to”, 24+ instances, “she forced herself to”, 11+ instances. 

The text also relies on a few physical descriptors that cycle repetitively. These mechanically staged reactions appear in literally painfully predictable patterns every 2-3 paragraphs during emotional scenes:

Throat (14+ instances):

“Her throat went dry” (3x)

“throat working” (2x)

“throat went tight” (2x)

“throat catching”

“throat worked, a visible swallow”

Heat/temperature (18+ instances):

“Heat crawled up Samira’s face”

“heat crept up the back of her neck”

“heat rising in her chest”

“cheeks heating”

“Heat crept up her neck”

“skin warming in the small closet”

Pulse/heart (11+ instances):

“Her pulse spiked”

“pulse ticked faster”

“pulse thrumming”

“pulse jump traitorously”

“Her pulse climbed”

“heart was hammering”

Stomach (9+ instances):

“stomach twist”

“stomach flipped”

“stomach dropped”

“stomach turned”

“stomach bottomed out”

Jaw/teeth (8+ instances):

“jaw clenched”

“jaw flexed”

“jaw tightened”

“teeth grind”

“grinding nature”

The [dialogue] > [tag with adverb] > [character gesture] > [internal reaction] pattern repeats with minimal variation throughout the almost 40,000 words. Characters also perform the same gestures repeatedly during conversation!

“crossed arms” (7x)

“tilted head” (11x)

“arched/raised eyebrow” (8x)

“jaw clenched/flexed” (11x)

Now, this was my favorite bit. This is where the uncanny valley, like, really lives. Because you’re reading, and reading, and reading, and then… Fuck, pattern recognition is a hell of a drug! Many paragraphs follow a template of [Physical reaction], [Question] and/or [Realization] and [Interpretation]. Remarkable consistency here. I’ve provided examples, because I’m a very sweet person that way.

Example 1: Her pulse spiked. [Physical reaction] This was her opponent? [Question] She had no idea who the woman was. [Realization] Something about that alone prickled under Samira’s skin. [Interpretation]

Example 2: Samira’s throat went dry. [Physical reaction] Was that a dig? [Question] Was she being mocked? [Question] This woman did not know her. [Realization] How could she possibly have expectations about her play three moves in? [Interpretation]

Example 3: Her stomach twisted. [Physical reaction] What was with chess today? [Question] First this Robby guy, and now her boss rattling off about a charity tournament. [Realization] The overlap made her head spin. [Interpretation]

Example 4: Heat crept up her neck. [Physical reaction] Who the hell did she think she was? [Question] She was fifteen minutes late and still in no rush to start. [Realization] Intimidation prickled down her spine, but annoyance followed right behind. [Interpretation]

Example 5: Her chest tightened. [Physical reaction] Why were they all staring? [Question] It was one table in a sea of tables. [Realization] She told herself they would move on, but the crowd did not thin. [Interpretation]

Example 6: Samira’s pulse climbed. [Physical reaction] A grandmaster. Whatever that meant. [Question/Realization] All she knew was she had been shoved into the seat because her boss’s martinis had won the night before. [Realization] Now she was about to be thrown to someone who would have no idea she had only ever played her father in a basement and strangers in the park. [Interpretation]

Example 7: Emery’s stomach dropped. [Physical reaction] What was happening? [Question] She had not felt this disheveled or shaken in years. [Realization] She was a mess. [Interpretation]

Example 8: Emery’s grip tightened on the wheel. [Physical reaction] Where had Samira been? [Question] In Brazil, at least, there had been Norhav. [Realization] Beyond that, they were all dull and predictable. [Interpretation]

Example 9: Her hands were trembling. [Physical reaction] What was she doing? [Question] She hadn’t felt this disheveled or shaken in years. [Realization] She was a mess. [Interpretation]

Example 10: Samira’s breath caught. [Physical reaction] Professional chess. What a joke. [Reaction] Her fingers brushed over the pair of business cards tucked in her pocket. [Action] She shoved the thought down hard. [Interpretation]

Isn’t it crazy to see all of that laid out? When you can, like, really see it? A lot of paragraphs also follow a 3-4 sentence template of [Action], [Interpretation] and/or [Elaboration], then [Result]. Examples are as follows:

Example 1: Samira fought the urge to flip the board in front of her as the man across from her glanced down at his watch for the third time in ten minutes. [Action] She knew the type. [Interpretation] Men who came down to Washington Square Park on their lunch breaks from their soul-sucking corporate jobs, desperate to prove something to themselves by beating a stranger at chess. [Elaboration] Every time, they left Samira’s table with mumbled insults and half-hearted jabs. [Result]

Example 2: She felt it before she saw it, someone watching her from afar. [Action] It was not in the usual way, either. [Interpretation] Most spectators came and went, nodding in mild approval or snapping a photo before moving on. [Elaboration] This man lingered. [Result]

Example 3: Samira froze as the wind cut across the square again, icy on her face. [Action] She could still hear the sound of traffic passing by and the rustle of chess clocks being slapped a few tables down. [Interpretation] It all felt muffled now. [Result]

Example 4: Her pulse spiked as the questions blurred in her head. [Action] They crowded the table, shoving microphones in her face while leaning over the toppled pieces. [Elaboration] One man’s hand closed around her arm to turn her for a picture, and she flinched so hard she shot up from her chair. [Result]

Example 5: Emery gripped the wheel of her Mercedes with one hand, the other drumming against her thigh as she threaded her way through Manhattan traffic. [Action] Heather had sent a driver, the usual bullshit song and dance, but Emery had waved him off without a second thought. [Elaboration] If the fundraiser turned out as dreadful as she expected, she wanted the freedom to slip away without explanation. [Result]

Example 6: Heat crawled up Samira’s face, but she shoved it down hard, leaning forward to nudge her bishop into play. [Action] The crowd stirred louder than before, their whispers rising to more steady conversation. [Elaboration] She got the strange sense they were talking about her. [Result]

Example 7: A horn blared a few blocks away as a dog barked. [Action] Somewhere behind her, a child screamed while his parents tossed him in the air. [Elaboration] Samira leaned back in her chair and let the noise of the city drown out the part of her brain that wanted to feel something. [Result]

Example 8: Samira hesitated, but stepped forward to take them. [Action] Her guard cracked just slightly at the mention of another woman, Dana. [Interpretation] That alone disarmed her more than she expected. [Result]

Example 9: The cardstock was thick and clean, the kind that meant someone had spent real money on it. [Action] She flipped one over in her hand, eyes scanning the lines. [Elaboration] There was a tactile reality to it. [Interpretation] These weren’t fakes or a common scam she saw all over the city. [Result]

These structures repeat across hundreds of paragraphs.

What makes this an issue is not any single repetitive pattern in isolation, it’s, and I’ll gladly repeat myself, literally the convergence and density of multiple formulaic structures occurring simultaneously across 40,000 words. Consider the mathematical fucking improbability: a text that maintains identical sentence complexity across two distinct POV characters, while also cycling through the same 8-10 emotional descriptors every 2-3 paragraphs, while also adhering to paragraph templates, while also deploying the exact same 12 physical reaction cues in predictable rotation, while also inserting hedging qualifiers at consistent intervals, while also filtering every observation through perception verbs. This literally just defies organic human composition. 

Human writers inevitably introduce variance. We don’t naturally maintain algorithmic consistency across tens of thousands of words because our attention, energy, and creative choices fluctuate. The probability that a human writer would unconsciously maintain this level of pattern fidelity is at a zero. LLM models, however, maintain stable patterns because they’re optimizing for coherence within learned parameters. They are not creating from embodied human experience. And they’re only, unfortunately, getting better, which means con artist writers will just get away with it and continue their cloutfarming streak with a little pep in their walk.

Just to preface, before we start:

COLLIDE:

Published: 2025-04-28

Updated: 2025-07-18

Words: 170,624

UNSCRIPTED:

Published: 2025-08-02

Updated: 2026-02-10

Words: 119,067

And general sources I always have on hand regarding AI-generated text vs. real human writing: 1 2 3 4

Let’s go!

The excerpt I looked at, which is only Chapter 1 of “COLLIDE”, generally exhibits a very, very dense cluster of features (repetition, formulaic structure, vague/generic description, grammatical/tonal over-standardization, lack of interiority or authentic perspective, and language that keeps rephrasing the premise) that literary editors, teachers, and stylometric research repeatedly associate with AI-assisted prose.

AI-assisted fiction often reads as if the model was given a premise and told to expand it attractively. That produces prose that keeps restating the same setup in fresher-looking language, like a rewrite. I call this redundant exposition, or in this case, summary contamination; the synopsis leaks into the prose everywhere! When someone prompts a model with summary, tropes, and chapter/story goal, the resulting text often retains the setup material.

Consider this:

“Your world is all polish and perfection, hers is chaos and self-destruction”

“[...] polished gold and tarnished silver, bright spotlight and flickering neon”

“Your voice soars in pristine arenas, hers snarls through smoke-filled stadiums”

“The perfect pop princess and the self-destructive rockstar”

“One is soaring, the other is spiraling”

“One has everything to prove, the other has nothing left to lose”

“The golden rise and the inevitable fall”

“an unstoppable force crashes into an immovable object”

Eight distinct antithetical constructions in roughly 150 words of prose. That’s dense as fuck! Like, I’m sure we get it!

Human writers use antithesis because it’s a classical rhetorical figure that has been around for-fucking-ever, but concentrating it this heavily produces a very unique textural effect that published fiction editors, for example, typically cut.

Instruction-based LLMs have been extensively reinforced to produce “literary” or “evocative” prose in response to requests for same, and antithetical parallelism is one of the highest-frequency devices they’ve learned to deploy. Muñoz-Ortiz et al.’s 2023 work on “Contrasting Linguistic Patterns in Human and LLM-Generated News Text” documented this elevated frequency of parallel oppositional structures as a reliable discriminator. Later work by Reinhart et al. (2025) similarly found broad grammatical, lexical, and rhetorical differences between human and LLM-generated text across genres.

Fanfiction does lean on contrast tropes, especially in the setup of two protagonists. However, even within fanfic, the density here is at the upper edge. A human fanfic writer working this trope would typically actually build scenes that demonstrate the contrast instead of shoving it down our throats like this. The premise is already fully packaged in the summary:

“A pop princess on the rise. A rockstar on the edge of collapse.”

“You need edge. She needs damage control.”

“one polished, one reckless”

And then the body of Chapter 1 keeps re-re-re-announcing that exact same contrast (see above)! That is not normal narrative repetition in the sense of real motif; just paraphrasing the premise over and over, sort of like it was repeatedly reading its own blurb and turning the blurb into more blurb. 

Next are the damning lists of three, with each item more abstract or intensified than the last:

“in her blood, in her bones, in the way her fingers itched”

“something raw, something electric, something different” — so, what?

“Gritty, raw, the kind of thing that made you feel something deep in your chest” — this, by the way, is a fat fucking tell, because what the fuck is the kind of thing that made you feel something deep in your chest? What the fuck is it?

“sweat on her skin, the burn in her throat, the raw, unfiltered power of the stage”

“Marketing plans. Endorsement deals. Feedback from the label.”

Tricolon is another classical writing device that goes way back (just take a look at Julius Caesar’s veni, vidi, vici!). It’s not suspicious on its own. But the frequency of it in LLM-outputs combined with a specific pattern: the third term in each tricolon tends to often pivot from something concretely named to abstraction (blood > bones > the way her fingers itched???; raw > electric > different???).

But this also brings another point to the conversation. One of the best editor-level heuristics is that AI often gives you the idea of specificity rather than something truly concrete. Wikipedia’s AI-writing guide and commentary on it, which I linked above, describe this as LLMs sanding down the unusual and nuanced into a broader, more generic and generally statistically common description. Writers and teachers also keep describing machine prose as uniform and low in interiority. That is all over this excerpt.

The vocabulary is saturated with abstract “packaging” words, which are all extremely common in AI fanfiction. Such as:

fame

timing

spotlight

perfection

chaos

self-destruction

edge

obsession

story

destiny

Those words are not wrong, real writers use them all the time; the problem is their density in LLM prose! Density and convergence of all of the signs I talk about in my other posts are the key to checking whether a fanfiction may be AI or not.

Moving on, supposedly, this is a whole music-industry world, yet the details are mostly generic placeholders:

“marketing plans”

“endorsement deals”

“the label”

“Billboard”

“The Voice”

“TMZ”

“sold-out tour”

Recognizable mumbo-jumbo cultural wallpaper, anyoneee? Likewise, the performance writing is generic in a way AI oh-so-often is:

“The first note hit like a punch to the chest”

“Dina’s bassline crawled under the melody”

“the air felt electric”

Nothing there is impossible for a human to write. The issue is that it’s fucking AI slop, because if you prompted ChatGPT to write something like this, lo and behold, you’d catch the exact same phrases, the exact same words, expressions, similes, and so on.

This chapter also massively overexplains character psychology and the general theme, almost like a complete caricature. One of the most repeated complaints is that AI prose does not trust implication whatsoever. It states, glosses, translates, and re-translates, all using extremely vague language, repetitive structure, unnecessary exposition… well, I’m sure you’re starting to get the gist of it. This chapter pretty much never lets the reader infer. It’s just so goddamn pre-chewed. Even the closing section is not really narrative closure. It’s a fucking voice-over summary!

Another thing worth noting, is the copyright notice they slap everywhere, which goes as follows:

© VALEISASLUT, 2025. please do not repost, translate, copy, or distribute without permission.

The stupid version of the argument is that you can’t copyright fanfiction at all. Technically, fanfiction is a derivative work. Under U.S. copyright guidance, derivative works can contain copyrightable new (!) material, but that protection extends only to the writer’s original additions, not to the underlying canon they borrowed from. In other words, nobody suddenly owns Ellie Williams, Joel Miller, The Last of Us, or the franchise’s commercial universe because they wrote a coffee shop AU about them, obviously. But you may still own your specific wording and other genuinely original expression you contributed.

AO3’s own policy reflects that! OTW’s legal position is that most noncommercial fanworks are transformative and usually fair use, and AO3’s TOS states that complaints based merely on the existence of fanwork using copyrighted material will not be pursued. At the same time, AO3 is crystal clear that plagiarism is a violation, unauthorized reposts require permission, and translations and podfics also require permission from the fanwork creator if AO3 is going to host them.

The author cannot claim a monopoly over canon material they do not own by using a decorative ©. The fanfic itself is only able to exist because it is borrowing an entire preexisting commercial universe, so it’s also rhetorically just inflated. She can say “do not repost my exact text” or “do not translate this without permission”, but the aura it tries to project is that this is some kind of sealed, proprietary object in the full legal sense, when in reality it is a derivative, transformative work whose “protectable” core is limited to whatever original expression the author actually added—and, if we consider that this is AI-generated, then that copyright notice is lunatic.

Because, under the US Copyright Office’s January 2025 guidance (Part 2 of the Copyright and AI Report), AI-generated content is not copyrightable at all. Human authorship is a “bedrock requirement”, affirmed by the DC Circuit in Thaler v. Perlmutter. Works with mixed AI/human content require the AI portions to be disclaimed. So if this person is using AI significantly, which I strongly believe they are, and claiming blanket copyright, they’re literally in tension with current federal guidance. <3 Yay. <3

Moving on: the “proof” video. Dawg. Every fucking day on this app, man, I swear…

Here is the caption attached to it:

“come with me to write unscripted chapter 8! this one took multiple sessions across 7 days, but i managed to make it a 3 min video. huge thank you to nini and andrea for helping me proofread it was actually so fun and i should do this way more often. keep throwing those false accusations if you want clearly it just gives me motivation to keep writing and making these little videos💜💜💜💜”

As evidence of authorship, that is non-probative. The video absolutely fails as a rebuttal. A cropped, sped-up shot of a Google Docs window does not establish original composition! Oh my Goooddd. It establishes only that text was entered into a document. If the visible typing is linear, which it very much seems to be, and more like transcription than actual composition, then the video can be read just as easily as transcribing from notes, retyping from another monitor, from a phone! It does not show messy composition whatsoever, or real drafting, hesitation, backtracking, deletion, substitution, sentence restarts, structural movement, paragraph shuffling, open research tabs, abandoned phrasing, tone recalibration… I could go on and on. I watched that fucking video and it’s literally crazy that this was posted in “good faith”. It is evidentiary theater, but no evidence!

At best, it proves that somebody typed text into Google Docs and had friends proofread it afterward. If they wanted to make a true video to prove themselves, I’d only settle for a literal camera set up behind their head, with absolutely no potential of transcription from anything around them. But, see, that sounds insane, and I don’t believe them for shit anyway, so I don’t need that.