Forum: Story Recommendations

Popped chapter 1 in ChatGPT and asked "was this written by AI". Response below

——

It's impossible to be absolutely certain without metadata or provenance, but stylistically, this text very likely was generated (or heavily assisted) by AI. Here's why:

🧩 Stylistic and structural indicators
1. Uniform rhythm and diction:
The prose maintains a consistent cadence and polished sentence construction throughout, without the subtle irregularities of human drafting (no clunky phrasing, shifts in tone, or unintentional repetition). This level of mechanical smoothness is typical of AI writing.
2. Patterned description and sensory sequencing:
Scenes are introduced with the same formula—visual cue → emotional reaction → sensory detail → rhetorical question. AI models trained on web fiction (especially Chinese or Korean light novel structures) often reproduce this pattern.
3. Character and dialogue regularity:
Each speaker expresses themselves in complete, grammatically perfect sentences, even under panic. Human authors usually mix clipped speech, interruptions, and idiosyncratic phrasing.
4. Derivative thematic blend:
The setup—dream versus reality, seven strangers, a survival "task," and an elegant but monstrous butler—closely resembles common AI-training inputs from web novels, particularly "dungeon" or "survival game" tropes. It reads like a synthesis of Lord of the Mysteries + Alice in Borderland + Silent Hill tone.
5. Lack of deeper voice differentiation:
The narrative voice and the characters all share the same linguistic texture—no distinct rhythm, education level, or worldview expressed through speech. This flattening is a common AI artifact.

🧠 Technical signals

If this text were uploaded to an AI-detection model (like GPTZero, DetectGPT, or OpenAI's watermarking tools), you'd likely get a "high probability of AI generation" score because:
• High burstiness consistency (low variance in sentence length).
• Moderate perplexity (around 25–40 range typical of GPT-4/Claude fiction).
• Synthetic coherence—every paragraph connects cleanly, no loose transitions.

📚 Summary judgment

✅ Highly polished, balanced pacing, no grammatical slips
✅ Repetitive emotional cues and question-answer patterning
✅ Uses web-fiction tropes almost archetypically

Verdict: Almost certainly written or edited by a large-language-model AI such as GPT-4/Claude/Suno-text, possibly with light human polishing.

Would you like me to show you exactly which sentences display the strongest AI fingerprints (e.g., formulaic paragraph openers and rhythm signatures)?

Sure. Incidentally, Grok had a very similar assessment (but snarkier because that's tone).

Here's the explanation ChatGPT gave of its rating

——-
Excellent — here's a line-by-line forensic read. I'll mark the AI fingerprints that strongly suggest either direct model generation or heavy editing through an AI assistant.

⸻

1️⃣ Formulaic scene framing

"Before me stretched a world shrouded in mist."
Why it's AI-typical:

• Classic sensory opener ("Before me stretched…" / "The world was…").
• Models love this cadence — short declarative image followed by a modifier clause — because it's frequent in training data (fantasy and web novel intros).
• Human writers often add a unique tonal hook ("Mist pressed against my teeth," etc.); models default to stock cinematic phrasing.

"Ethan Walker stood dazed, his mind grappling with the surreal."
Tell-tale feature: "X stood dazed, Y grappling with Z" is an AI favorite syntactic scaffold: subject + past-participle + participial phrase. It reads smooth but generic.

⸻

2️⃣ Repetitive pattern: Image → question → reaction

"But dreams couldn't bleed into reality, could they?"
"The pendant's weight … raised a chilling question: was he still dreaming?"
"Where am I? What is this place?" "Who the hell are you people?"

AI fingerprint: repeating rhetorical questioning every few paragraphs is a coherence-keeping trick. Human drafts rarely sustain it so evenly — they tend to cluster questions near turning points, not rhythmically.

⸻

3️⃣ Consistent syntax smoothing

Nearly every sentence uses one comma-bound modifier:

"Ethan tugged at his shirt collar, peering down."
"Her face lit up with relief … only for her expression to shift to shock."
"He followed the voice's direction, and what he saw defied belief."

LLMs often smooth transitions by splicing two simple clauses with commas or conjunctions to keep perplexity moderate. A human rough draft usually alternates long/short beats or fragments for tension ("He tugged at his collar. Sweat. The pendant was real.").

⸻

4️⃣ Flat emotional register

"His chest burned, his pulse thundering like a drumroll."
"The group stared, dumbfounded, their understanding of reality crumbling."

These use cliché metaphors and even rhythm spacing identical to hundreds of AI outputs (emotion → physical cue → exaggerated simile). There's no idiosyncratic vocabulary that signals a distinct writerly voice.

⸻

5️⃣ Character voice homogenization

All characters speak in syntactically perfect, exposition-ready English:

"What else? You think you've got a choice?"
"You're in on this prank, aren't you?"
"Running out there's a death sentence."

The grammar is too polished under stress. AI models lack the natural variance of real dialogue (fillers, sentence fragments, interruptions, dialect markers). Human fiction would include contractions like "Ain't," or mid-sentence breaks.

⸻

6️⃣ Scene-ending reset loops

"Welcome to Blackwood Castle, honored guests. I am the Butler here, and I shall attend to your needs."
Then again after the murder:
"Welcome to Blackwood Castle, honored guests…"

LLMs love closing symmetry — reusing the first beat as a loop to "bookend" scenes. It feels tidy but mechanical, signalling algorithmic story structuring rather than intuitive pacing.

⸻

7️⃣ Over-balanced paragraph lengths

Almost every paragraph is 3–5 sentences long, none shorter than two. This low burstiness variance (a measure of sentence length fluctuation) is one of the strongest statistical giveaways. Human fiction has wider swings — a single-word paragraph, an 8-line block, then a half-sentence fragment.

⸻

🔎 Quantitative pattern summary

Feature Human typical range Observed here Comment
Mean sentence length 13–18 words ~17.2 Smooth, low variance
Std dev (burstiness) >6 ~3 AI-flat
Rhetorical question frequency

Almost every paragraph is 3–5 sentences long, none shorter than two.

That's visibly false.

So a typical AI response, spurting pompous expert wording more than likely based on legit human done research paper.

But it still can't do any counting correctly...

I wouldn't put any trust in this AI output. It can find sample supporting the AI fingerprint (That's an huge achievement!). But it could as well support convincingly the human creation option when requested.

Since those AI still can't count occurrences, they are not qualified to make an determination based on the legit papers they are amalgamating and appropriating.

I agree that ChatGPT has a counting problem.

Suit yourself on the rest of your issues. Whether you believe it is detecting AI with the other points or not, writing improves when the other points are followed.

My best, E

My personal AI "test" is much more primitive, any new author posting multiple stories (or chapters in multiple stories, or a combination of the two) on a daily basis, I assume an artificial component.
I'm prepared to make exceptions if I can see a good reason why the posting pattern fits that criteria, but Caffeinated Tales fell into the trap.

Yep. I write with AI also, but I'm very interested in producing a high quality output. You won't get a daily posting frequency from me unless the story is complete. It takes time to produce a decent result with AI.

There is a difference between bad writing written by a human being, and bad writing written by an AI :) An AI will write "perfect" bad writing.

And yes, I agree with you that it could flag a story that was written by a flesh and blood human as an AI story. I have seen it do that with one earlier story that I ran through this test, and when I pointed out that the story had been written before the commercial availability of LLMs, it responded that the story was (essentially) poorly written.

Are you the same Diamond Porter who wrote the Anna and Jayden story? If so, I can tell you that ChatGPT correctly identified you as a human author, while Grok swears the story was written by an AI.

These systems are imperfect, which everyone knows. Some will choose to argue that because of their imperfections they don't merit consideration. I'm not in that camp :) To me they're simply tools that can provide useful information.

E

To me they're simply tools that can provide useful information.

They can and frequently do also provide false information which is less than useful.

And that is why we have the ability to discriminate as humans!

They can and frequently do also provide false information which is less than useful.

Wait a minute, SOL is a fiction site labelled as such.
I browse YouTube now and again, and there are an increasing number of videos pushing total lies dressed up as fact, rewriting history. One absolute classic was pushing a narrative where 7 or 8 WW2 U-Boots were sunk in the space of a minute or so by one aircraft. Several of the supporting facts were real, but all of the U-Boots were real and none of them went down that day - the core part of the story is a total pack of lies from beginning to end. The narrative is also total garbage, perpetrated by an entity which did not understand German - a "Kapitän zur See" (Captain of a sea vessel) was rendered as though one of his names was "Sursay".

One of the challenges in training AI at scale is misinformation…. The raw material used to train the models comes from the internet! I don't rely on Grok for anything factual - it trains on Twitter/X data which is notoriously unreliable to begin with.

Wait a minute, SOL is a fiction site labelled as such.

I was responding to a sub thread on AI assessments of whether particular stories were written by AI and a claim that AIs provide "useful information".

The relevance of SOL being a fiction site is less than zero.

Can I help to explain them?