AI Flagging & False Positives – Anyone Else Seeing This?

[SkyBubble]

After 19 years, Literotica’s AI detection farce has me fed up. I submitted The Contest: Becoming Sarah (9,635 words, 100% human-written) and got an AI rejection despite ~0% on ZeroGPT (six sections). I can't keep jumping through these hoops, especially when I already did.

Since March, I have written and rewritten, worked with a Volunteer Editor, and run AI checks, having faced false flags on Homecoming Ch.03 and Gretchen’s Unexpected Passion (one took four submissions). Their rejection suggested an editor--which I found a little frustrating, since I already used one. Another story got approved in hours, unlike The Contest’s 11-day wait.

My 2006-2009 stories, pre-AI, scored 20-55% (e.g., The Deflowering of Rebecca at 46% minimum), and today’s re-tests show this system’s not working properly. I’m re-testing The Contest, for informational purposes. It’s thriving on AO3 (Sky_Bubble), Tales After Dark, Writing.com, Wattpad, and other sites. I tailored it for this site, but they rejected it anyway. It's frustrating. Fortunately, it's doing extremely well elsewhere.

As I mentioned, this is the third time I've been hit with a false flag. I'm getting very close to taking down my stories and leaving. I'd prefer not to do that, but I don't have any more time to play these games.

Anyone hit by AI rejections? Run pre-AI stories through ZeroGPT/humanizer--false positives? If Literotica’s response isn’t satisfactory, I’m pulling all my stories and leaving this mess. Thoughts?

 

[NotWise]

After 19 years, Literotica’s AI detection farce has me fed up. I’m resubmitting The Contest: Becoming Sarah (9,635 words, 100% human-written) from Drafts after an AI rejection despite ~0% on ZeroGPT (six sections).

Since March, I have written and rewritten, worked with a Volunteer Editor, and run AI checks, having faced false flags on Homecoming Ch.03 and Gretchen’s Unexpected Passion (one took four submissions). Their rejection suggested an editor—insulting. Another story got approved in hours, unlike The Contest’s 11-day wait.

My 2006–2009 stories, pre-AI, scored 20–55% (e.g., The Deflowering of Rebecca at 46% minimum), and today’s re-tests [pending: e.g., “varied 10–15%”] show this system’s unfair. I’m re-testing The Contest [pending: e.g., “flagged minimal issues, ~0% AI”]. It’s thriving on AO3 (Sky_Bubble), Tales After Dark, Writing.com, Wattpad, and other sites.

Anyone hit by AI rejections? Run pre-AI stories through ZeroGPT/humanizer—false positives? Lush Stories rejected it for grammar (articles/pronouns, adverbs)—tips? If Literotica’s response isn’t satisfactory, I’m pulling all my stories and leaving this mess. Thoughts?

It's a common complaint.

You should know that AI detectors may only detect the work coming from one AI, so your "0% on ZeroGPT" may be higher with a different AI detector. Something that also surprises people is that Grammarly's suggestions are from a generative AI.

If all else fails, then there's always the other platforms you mentioned.

 

[femmeappeal]

Something that also surprises people is that Grammarly's suggestions are from a generative AI.

It won't just be Grammarly, not by a mile. Any system that claims to be using AI (it's not obviously) to help you improve a post, comment, article, or document is using a generative LLM to accomplish the task. And therefore any of those tools can get you a ping from something trying to detect LLM writing.

Ironically, if there was such a thing as actual Artificial Intelligence, it would be a lot easier for detectors to do what they do, without false positives. But because it doesn't exist (and we have no idea at this point if it ever will) and generative LLMs do not qualify as actual computer-based intelligence, the detectors have a limited ability to do anything except look for output that's specific to specific generative tools.

 

[BobbyBrandt]

Try submitting it as an uploaded file instead of pasted into the text field. Let us know the results.

 

[NotWise]

Try submitting it as an uploaded file instead of pasted into the text field. Let us know the results.

Why would that make a difference?

 

[ElectricBlue]

Anyone hit by AI rejections? Run pre-AI stories through ZeroGPT/humanizer—false positives? Lush Stories rejected it for grammar (articles/pronouns, adverbs)—tips? If Literotica’s response isn’t satisfactory, I’m pulling all my stories and leaving this mess. Thoughts?

It's heresy saying this, but there must be something in your style of writing that's triggering AI detectors, if your content is triggering so many (including whatever it is that the site uses). They're all a crap shoot, sure, but that might not be the issue.

I don't know what AI detectors do to detect AI content, but there's a "flavour" to generated content that simply doesn't "feel right". You can't put your finger on it easily, it's not always consistent, but it's there - I see it often in very bland business reports. The type of report someone has run through their grammar checker, but not paid much attention to how it actually reads.

This probably not very helpful, and you're not the first to get walloped for suspected use of AI, but you must be doing something that most of us here don't do, because our stories are getting through, getting published. And presumably our content is going through the same process.

 

[ShelbyDawn57]

It's heresy saying this, but there must be something in your style of writing that's triggering AI detectors, if your content is triggering so many (including whatever it is that the site uses). They're all a crap shoot, sure, but that might not be the issue.

I don't know what AI detectors do to detect AI content, but there's a "flavour" to generated content that simply doesn't "feel right". You can't put your finger on it easily, it's not always consistent, but it's there - I see it often in very bland business reports. The type of report someone has run through their grammar checker, but not paid much attention to how it actually reads.

This probably not very helpful, and you're not the first to get walloped for suspected use of AI, but you must be doing something that most of us here don't do, because our stories are getting through, getting published. And presumably our content is going through the same process.

I agree. Not that it's heresy, but that AI detectors are style checks. Basic logic says that since you have no control over the AI check but you do have complete control over your style, you can either enhance(it's a more forgiving word that change) your style, or continue to get rejected. Right or wrong, it's just the way it is.

Vary sentence length. Change up dialog tags. Get creative with word choice; a thesaurus is not an IA(in most cases). Show, don't tell. You've been writing here for 19 years +/-, you've heard all this before. Be creative. I wish there were a formula...

Good luck

 

[AwkwardMD]

It's heresy saying this, but there must be something in your style of writing that's triggering AI detectors, if your content is triggering so many (including whatever it is that the site uses).

It doesn't need to be something in their style. He could be using leaning on Grammarly, or Co-Pilot, and not wanting to own up to it.

I don't know how many of you have followed this users posts over the past 6 months, but I've replied to him a number of times and gotten little in the way of meaningful conversation. Just a dogged, dogmatic recitation of this many rejections of this and that many rejections of that. Even this post here is Skybubble saying "I'm about to submit a story" after making different thread saying "I'm taking my ball and going home" within hours. Most of us have been here long enough to remember someone making their "I'm leaving" and then coming back, but how many times did it happen this quickly?

Methinks something something protesting something something.

 

[VerbalAbuse]

Vary sentence length. Change up dialog tags. Get creative with word choice

that's exactly what the AI does and many writers don't.

 

[AwkwardMD]

If the AI detector was this fragile, this prone to false positives, more of the rest of us would be experiencing periodic rejections.

We are not.

 

[ShelbyDawn57]

It doesn't need to be something in their style. He could be using leaning on Grammarly, or Co-Pilot, and not wanting to own up to it.

I don't know how many of you have followed this users posts over the past 6 months, but I've replied to him a number of times and gotten little in the way of meaningful conversation. Just a dogged, dogmatic recitation of this many rejections of this and that many rejections of that. Even this post here is Skybubble saying "I'm about to submit a story" after making different thread saying "I'm taking my ball and going home." Most of us have been here long enough to remember someone making their "I'm leaving" and then coming back, but how many times did it happen this quickly?

Methinks something something protesting something something.

I know you're not replying to me, but I agree with @AwkwardMD.

I've had this conversation with a few others. Two of them sent me snippets of their writing. Scanning them, I got high percentage of AI likelihood. With a few stylistic modifications, I was able to get both to pass as human 100%. I'm basing my opinion on AI check being a style check on that experience.

 

[ShelbyDawn57]

that's exactly what the AI does and many writers don't.

Not in my experience. YMMV...

 

[AwkwardMD]

My 2006–2009 stories, pre-AI, scored 20–55% (e.g., The Deflowering of Rebecca at 46% minimum), and today’s re-tests [pending: e.g., “varied 10–15%”] show this system’s unfair. I’m re-testing The Contest [pending: e.g., “flagged minimal issues, ~0% AI”]. It’s thriving on AO3 (Sky_Bubble), Tales After Dark, Writing.com, Wattpad, and other sites.

Have a quick look at these three bolded sections. First person to spot the weirdness gets 5 internet points.

 

[filthytrancendence]

Have a quick look at these three bolded sections. First person to spot the weirdness gets 5 internet points.

Honestly most of that sentence is very weird. But I'm not sure exactly what you're referring to so I guess I miss out on 5 points

Is it a normal thing to mix parenthesis and square brackets in the same sentence and I just missed the memo? What are the square bracketed bits even meant to convey? Is there even a widely understood linguistic function to square brackets? I pretty much only use them to make an [insert joke here] joke...

 

[ShelbyDawn57]

Honestly most of that sentence is very weird. But I'm not sure exactly what you're referring to so I guess I miss out on 5 points

Is it a normal thing to mix parenthesis and square brackets in the same sentence and I just missed the memo? What are the square bracketed bits even meant to convey? Is there even a widely understood linguistic function to square brackets? I pretty much only use them to make an [insert joke here] joke...

Pretty sure what @AwkwardMD was pointing out is the implication that @SkyBubble scanned his older stories and found that they got rated between 20 - 55% possibility of being AI. I may be mis-inferring something, but that's what i took away too. Adding in the todays retest variation of 10-15% skews those scores on the high end as high as possibly 70%.

 

[filthytrancendence]

Pretty sure what @AwkwardMD was pointing out is the implication that @SkyBubble scanned his older stories and found that they got rated between 20 - 55% possibility of being AI. I may be mis-inferring something, but that's what i took away too. Adding in the todays retest variation of 10-15% skews those scores on the high end as high as possibly 70%.

The way I interpret the sentence is that his pre-AI stories are being flagged as AI, which he is saying should be impossible if the test is actually checking for AI generation. The %'s are laid out in a nearly incomprehensible way.

And I'm pretty sure that's the most generous possible interpretation of what he's saying.

There are a number of problems with this reasoning:

1. We don't even know what testing methodology the Lit mods are using, do we? So this is just a blind guess.
2. There isn't a way to test for AI generation. That's just a nonsense thing. Language is language. The best you could test for is patterns that AI generated text commonly follows. But there's no reason that a particular human's writing style couldn't mimic those patterns entirely incidentally. It seems to me there's enough variation in how people write that some 'false positives' are inevitable.
3. Even if it is a 'false positive', and even if the offender is being 100% honest about not using AI, all that means is that they accidentally mimicked the linguistic patterns of AI generated text on accident well enough to trick the testing algorithm into believing they are an AI. That's just a really complicated way of saying their writing is pretty bad.
4. Lord knows, pretty bad writing is not generally grounds to get a story rejected on Lit. But I have to wonder, what incentive does anyone running the site have to extend benefit of the doubt to people whose writing is pretty bad in the specific way that would cause their AI testing methodology to believe their text was AI generated?

 

[SkyBubble]

It's heresy saying this, but there must be something in your style of writing that's triggering AI detectors, if your content is triggering so many (including whatever it is that the site uses).

Yes, I deliberately unpolished this story, and it still triggered the system.

 

[SkyBubble]

Significant scores like that from an AI detector for stuff written well before all these AI tools shows how utterly unreliable they are.

 

[SkyBubble]

​

Most AI detectors, such as ZeroGPT, GPTZero, and Originality.ai, among others, claim success rates in the 60-70% range, at best. That means they label writing incorrectly at least 30-40% of the time, a huge margin of error--especially when used to judge creative work.

Why does this matter?

Detectors that misfire one third of the time aren't reliable enough to be used as gatekeepers for publishing, especially storytelling.
False positives disproportionately affect nuanced, polished, or emotionally subtle human writing.
Platforms that rely on these tools without human review or transparency erode trust with their authors.

Even the best tools struggle with older writing styles, poetic prose, and complex dialogue, so they often flag your own edited work as AI simply because you've polished it for clarity or rhythm, and can’t determine intent, emotion, or lived experience--just patterns and probabilities. The high AI probability rates on the older stories I ran through their system show this.

 

[StillStunned]

Leaving aside the eerie sense that your entire post feels like it's an AI reply...

• False positives disproportionately affect nuanced, polished, or emotionally subtle human writing.

And yes, even the best tools:​

• Struggle with older writing styles, poetic prose, and even complex character dialogue.
• Often flag your own edited work as AI simply because it's been polished for clarity or rhythm.

I polish the crap out of my stories. I consider every single word, by itself, in relation to the words around it and in the context of the paragraph, scene and story as a whole. I once wrote a 19k story entirely in the style of a 19th-century novel. I use all kinds of imagery and literary devices in my prose, including alliteration, repetition, consonance, onomatopoeias and rhythm.

None of my 63 stories have ever been flagged for possible use of AI.

So, it's clearly possible to publish stories here with advanced prose. If your style is triggering the detectors, perhaps you should see it as an opportunity to evolve your style.

Don't take that as an insult or offence: for centuries people have been telling writers to change their style if they want to get published. Publishers, critics or readers might expect something different from what the writer is offering. In this case, it's Laurel. It's her site, and if you want to keep publishing here she'd like you to adjust your style.

 

[Voyeurkenneth]

Here's the current landscape:​

Most public AI detectors (like ZeroGPT, GPTZero, and Originality.ai) claim success rates in the 60–70% range at best. That means they mislabel 30–40% of the time—a massive margin of error, especially when used to judge creative work.

Why this matters:​

• A detector that misfires one-third of the time is not reliable enough to be used as a gatekeeper for publishing.
• False positives disproportionately affect nuanced, polished, or emotionally subtle human writing.
• When platforms lean on these tools without human review or transparency, they erode trust with their authors.

And yes, even the best tools:​

• Struggle with older writing styles, poetic prose, and even complex character dialogue.
• Often flag your own edited work as AI simply because it's been polished for clarity or rhythm.
• Can’t determine intent, emotion, or lived experience—only patterns and probabilities.

This is clearly a ChatGPT response.

 

[BobbyBrandt]

Why would that make a difference?

Just a suggestion. There seems to be some evidence, although a small sample, that uploading a story that has been run through Grammarly or other AI tools see fewer rejections than when the same story is pasted into the text box.

Try it. Don't try it. Your choice.

 

[StillStunned]

If the AI detector was this fragile, this prone to false positives, more of the rest of us would be experiencing periodic rejections.

We are not.

It doesn't help your case if you also post things like this:

There are sites that are not so picky where you might get away with publishing an AI story.

Not that I'm suggesting you find ways to violate the rules, but did you run the story through and AI humanizer?

 

[AwkwardlySet]

 

[Actingup]

No, it hasn't happened to me. One other suggestion I have is to ensure that you turn off any predictive text on your word processor. AI is not 'intelligent' - it is just advanced predictive text, so the more your writing looks like something that a predictive text model would have expected to see, the more chance you get a false positive, she murmured, gently stroking down his spine. Including little surprises in the text might help too.