This is why we can’t have nice things

[SugarStorm]

Some credentials first. I work in IT, I like puzzles, and I happen to work with automatization. LLMs are part of my day job, and honestly, I quite like them — and I enjoy breaking them.

I’m going to show you how sensitive Lit’s AI detector really is. So if your story’s in Pending Purgatory, read on.

I write slow-burn lesbian romance, 30-40k words a piece, and I've been blessed with fast turnaround times when publishing. In September, Hearts Like Ours Pt. 01 went from submission to publication in 48 hours.

On Oct 10 I submitted Pt. 02, using same style, format and cinematic prose I enjoy writing. Crickets. After fifteen days, I pulled it. No AI rejection email — which I’ve faced twice before — just silence. Each time I re-uploaded, the submission timer reset within 24h, and then it got “stuck”.

I tried everything:

1. Delete/Resubmit
2. Used only official Lit-approved HTML tags
3. Uploaded docx instead of using the textbox
4. Removed formatting
5. Tried fewer tags
6. No em-dashes
7. Unticked the Series setting

I went through four different time-consuming iterations of my story with various stylistic tweaks. Nothing worked, and then finally, after a 2-3 weeks of tinkering, I gave up. Then two days ago (Nov 9): rejection for alleged AI use.

My vocabulary isn’t the best, I have a tendency to repeat myself and my love for em-dashes is second to none, but no — my writing’s my own.

So I tested it myself. I ran my original story through ChatGPT simulating AI detectors GPTZero, Copyleaks and Turnitin. ChatGPT can only handle 6-8k words due to its tokenization, so most of the story wasn’t even analyzed. Out of 34k words, only two paragraphs (early ones) were flagged for being “too polished”.

I then split the story into chunks I knew it could handle. Two more paragraphs were flagged, both mid-story: “dialogue too polished”.

Here’s the thing though: I don’t think Lit runs a full review. It likely samples, just like ChatGPT does when you give it a full sized document. In theory, enterprise customers can get more tokens (more money >> access to different models >> more tokens to work with) or pay for usage (costly), but I doubt Lit does either of those things.

Anyhow, I purposefully only modified the first flagged section to see what would happen next:

Original:

Riley Vaughn existed in a kind of limbo.

In the days since the breakup she hadn't been fully present. Not at work, not in meetings, and certainly not in herself. Her calendar still ruled her time, her inbox still dictated her mornings, and Mr. Jensen Ackerman, CEO of the bank and frequent source of migraines, still demanded her sharpest thinking.

But Riley's mind was somewhere else. Somewhere floral.

Sarah.

The sweet 25 year old running <i>Roses & Reverie</i> had imprinted herself like pollen on Riley's lungs, every breath thick with her memory. And Riley, for all her control and precision, hadn't figured out how to properly exhale since.

Revised:

Riley Vaughn existed in a kind of limbo. In the days since the breakup she hadn't been fully present. Not at work, not in meetings, and certainly not in herself. Her calendar still ruled her time, the inbox still dictated her mornings, and Mr. Jensen Ackerman, CEO of the bank and frequent source of migraines, demanded the best of her.

But Riley's mind was somewhere else. Somewhere floral.

Sarah.

The sweet twenty five year old running <i>Roses & Reverie</i> had imprinted herself like pollen on Riley's lungs, every breath thick with her memory. And Riley, for all her control and precision, hadn't figured out how to properly breathe since.

48 hours later, and with no submission timer being reset, Pt. 02 was published. So to summarize, I made four-five small changes on two paragraphs in a 105 page document and suddenly everything’s right as rain.

I will admit to having become a bit frustrated with the situation.

Here are my key takeaways, and best guesses:

1. The AI review seems to cover only the first 7k words or so of any story.
2. Lit’s AI detector is unreliable (whoosh). I doubt it’s home-built; more likely an API link to a commercial detector set to I hate you. Given that Lit already uses third-party load balancing, monitoring, and forum hosting, this fits the pattern.
3. If your timer resets, you’ve been flagged: for spelling, formatting, banned tags (don’t use “death”), requested edits, or possible AI content.
4. If you’re not flagged, I don’t think anyone at Lit ever reviews your work. That’s supported by how quickly large stories sometimes get approved.
5. This I think is the biggest change: the automatic AI rejection email we saw earlier this year has likely been moved into Literotica’s “review bucket", making it less visible and causing a major backlog or bottleneck. The “review bucket” has since become Pending Purgatory because Lit seem to review everything here manually.

So why do some stories go through after a delete/resubmit?

I'm guessing — probably luck. Just like ChatGPT can give you a slightly different answer to the same prompt, AI detectors can wobble at the threshold. If your story sits near that edge, sometimes you just get lucky and slip past it.

I get the need for an AI detector; I just don’t think this is it.

The next time your story ends up in Pending Purgatory, ask yourself if it’s really because it’s “too polished,” or just unlucky enough to confuse a fickle algorithm. The fun, apparently, is in the surprise.

And for that reason, I’m out. For a while, at least.

 

[AwkwardMD]

I can't imagine posting "Here's how I broke the AI detector and you can too" but also having the title "this is why we can't have nice things."

Honestly. What do you think you're actually doing?

 

[SugarStorm]

Honestly. What do you think you're actually doing?

I think the AI detector seems to be poorly implemented, it's gone from bad to worse, and I'd rather shine a light on the problem than not at all.

 

[AwkwardMD]

I think the AI detector seems to be poorly implemented, it's gone from bad to worse, and I'd rather shine a light on the problem than not at all.

Did I miss where your story was rejected for AI? I thought it said you were stuck in pending.

You don’t know that those are related.

 

[SugarStorm]

I went through four different time-consuming iterations of my story with various stylistic tweaks. Nothing worked, and then finally, after a 2-3 weeks of tinkering, I gave up. Then two days ago (Nov 9): rejection for alleged AI use.

 

[AmaraLust]

Honestly. What do you think you're actually doing?

Umm.. flexing the intelectual boner may be

 

[AwkwardMD]

You should redact that original post. It's wrong, but not wrong enough that it's fine and someone couldn't misuse it to make everything worse.

 

[SimonDoom]

I agree the post should be redacted or deleted.

More information isn't always better. Anybody who has run an organization knows this is the case.

 

[AwkwardMD]

Glitch AI rejections have happened in the past. SimonDoom had one. Posting how to beat the system is an unhinged reaction.

 

[gunhilltrain]

I think the AI detector seems to be poorly implemented, it's gone from bad to worse, and I'd rather shine a light on the problem than not at all.

I read Future Shock when it first came out. Now, decades later, I have some kind of "shock" almost every week. I don't have a copy any longer, but I'll have to see if the author had optimistic or pessimistic views. Literotica used to be a good website for writers. Now it's getting botched by this AI issue. I'm somewhat pessimistic myself.

 

[YmaOHyd]

I don't see this as posting about how to beat the system. It's just advice for getting through a process every single person seems to agree is frustratingly opaque and glitchy. Requests to redact the post seem... extreme.

I'll also say other people have posted, in great detail, about the automated tools they've used in their own writing and how to avoid flags, so the cat's already completely out of the bag if you choose to let Copilot or whatever pick adjectives for you.

 

[PennyThompson]

So I tested it myself. I ran my original story through ChatGPT simulating AI detectors GPTZero, Copyleaks and Turnitin. ChatGPT can only handle 6-8k words due to its tokenization, so most of the story wasn’t even analyzed. Out of 34k words, only two paragraphs (early ones) were flagged for being “too polished”.

I then split the story into chunks I knew it could handle. Two more paragraphs were flagged, both mid-story: “dialogue too polished”.

I don't want to wade into the entire situation and controversy here, but this part specifically is very wrong

ChatGPT is entirely incapable of "simulating" an AI detector. It's not designed for that function, it's not trained to do the kind of text analysis that AI detectors are trained to do. (Actual text analysis AI detectors have significant flaws, but that's a different topic)

When you asked it to pretend to be an AI detector on the text that you provided, all it did was fall back on its primary function as a text pattern replicator, it "knows" what it looks like to critique a text, so it makes up some realistic-looking nonsense that it thinks you want to hear.

This is like... hmm... like talking to a First Grader about how your car sounds squeaky, and then asking them if they think its brake pads need to be replaced. The kid can say "Yeah, I think your brake pads need to be replaced," which might be either correct or incorrect, but not because the kid knows what they're talking about.

 

[AwkwardMD]

It's just advice for getting through a process every single person seems to agree is frustratingly opaque and glitchy.

It's not every single person. That's just bias. The AH represents a tiny minority of the writers who actually use this site, and most of them seem to be posting just fine.

Advice on how to game a system is dangerous to share publically. Nobody wins if it gets worse.

 

[SugarStorm]

When you asked it to pretend to be an AI detector on the text that you provided, all it did was fall back on its primary function as a text pattern replicator, it "knows" what it looks like to critique a text, so it makes up some realistic-looking nonsense that it thinks you want to hear.

This is not actually what I asked though. I'm not giving out my prompt.

I will absolutely admit to not fully knowing LLMs in-depth, nor do I command AI detectors, but the realistic looking nonsense worked.

 

[Five_Inch_Heels]

Some...

First version seems clumsy and a bit hard to read. Second version flows better.

 

[YmaOHyd]

It's not every single person. That's just bias. The AH represents a tiny minority of the writers who actually use this site, and most of them seem to be posting just fine.

Advice on how to game a system is dangerous to share publically. Nobody wins if it gets worse.

No one's "gaming the system" here.

 

[PennyThompson]

I will absolutely admit to not fully knowing LLMs in-depth, nor do I command AI detectors, but the realistic looking nonsense worked.

I don't think it did, though. I think the first grader happened to say a thing in this one instance that happened to not contradict an unrelated thing that was occurring, and you connected the two.

I think this advice is like telling people to stop taking their car to the mechanic and start taking it to the local grade school

 

[SugarStorm]

Please redact or delete your original post.

I, and many others, spend a lot of time crafting our stories. Getting a simple rejection notice, with absolutely no worthwhile information, is not providing a level playing field. Lit can do more for its authors here.

I don't think it did, though. I think the first grader happened to say a thing in this one instance that happened to not contradict an unrelated thing that was occurring, and you connected the two.

It's a fairly big story though. You think it's down to chance it worked?

 

[AwkwardMD]

I, and many others, spend a lot of time crafting our stories. Getting a simple rejection notice, with absolutely no worthwhile information, is not providing a level playing field. Lit can do more for its authors here.


It's a fairly big story though. You think it's down to chance it worked?

I have been here the entire time. You and I are not unacquainted. You could have asked.

 

[YmaOHyd]

It's a fairly big story though. You think it's down to chance it worked?

I think your theory is sound -- limited chunks are fed into a free or cheap detector and rejections are handed out using those results. Your corrective action was right. Most AI detection sites operate on limited text; you can't drop a hundred thousand words into most of them. And you knew that you'd gotten an AI strike. So I don't think it mattered to the process that you checked it with ChatGPT and got plausible nonsense from the plausible nonsense generator.

 

[AwkwardMD]

Please delete or redact your post.

 

[EmilyMiller]

None of these threads, absolutely none of them, would be necessary if someone at Literotica would just tell us what the fuck is going on. That’s not a lot to ask, even as peasant supplicants to our feudal overlords.

 

[AwkwardMD]

I, and many others, spend a lot of time crafting our stories. Getting a simple rejection notice, with absolutely no worthwhile information, is not providing a level playing field. Lit can do more for its authors here.


It's a fairly big story though. You think it's down to chance it worked?

No response as to why you didn't approach me? It's pretty hard to miss the flag I wave on this subject.

 

[SugarStorm]

No response as to why you didn't approach me? It's pretty hard to miss the flag I wave on this subject.

I think you’re a Good Samaritan when it comes to AI rejections, but I don’t see you as the authority on the matter. For me, this was about more than just getting a story past the “approval” line.

Something needs to change. There’s no rhyme or reason to how things work right now. We’re asking for scraps, and not even getting that. The frustration is real.

I’ve shared what I’ve done and what I believe points to the issue, but I haven’t provided step-by-step instructions. The prompt I used stays with me.

If what I’ve posted is correct, then I suppose Team Literotica might delete the thread. Worst case, they remove my account and my stories along with it.

If it’s mumbo-jumbo, as PennyT suggested, then no harm done. It’ll just fade away like most threads do.

I’ve extended an olive branch to others who are struggling, and I’m genuinely curious to see whether I stumbled onto something real, or if it was just a one-off fluke.

 

[SugarStorm]

This silence undermines the "I tried everything" pillar this is all based on.

Two small humans here who need dinner and sleep first. Can't be everywhere at once.