Can AI Tell a Good Story?

I co-wrote a flash fiction piece with seven LLMs. Can you tell which one is me?

When it comes to AI writing, everyone has strong opinions.

“AI writing is easy to spot!”

“AI is better at writing than most humans!”

“AI can’t write good fiction!”

“AI doesn’t love you like we do, Dad. Please step away from the screen and come down from the attic. You haven’t eaten in days!”

I’ve heard it all.

In reality, things are always more nuanced than catchy headlines would have you believe.

As I’ve shown last year, while most of us can spot low-effort, generic AI slop, it’s easy to nudge large language models into tripping up AI detectors. AI can even be genuinely funny under the right conditions.

But can you, dear reader, easily tell which parts of a fictional story have been written by AI?

We’re about to find out!

The test

One of my Sunday Showdowns had two LLMs writing two separate microfiction pieces.

For this experiment, I’ve tweaked and expanded the premise and thrown in a human. Today, I’ll co-write a flash fiction story with seven free large language models.

Let’s see if you can easily tell which parts are AI-written.

The contestants

I wanted to involve a range of free LLMs, so here’s our final list:

1. Claude 4 Sonnet (Anthropic)

2. Daniel Nest (Why Try AI) - a real human person (allegedly)

3. DeepSeek-R1 (DeepSeek)

4. Gemini 2.5 Pro (Google)

5. GPT-4o (OpenAI)

6. Grok 3 (xAI)

7. Llama 4 Maverick (Meta AI)

8. Pixtral Large (Mistral AI)

The setup

The logline for our flash fiction story is:

In a future where all jobs have been fully automated, a human applies for the only position still open to biological applicants: "Empathy Calibration Technician."

Each of us will contribute a single segment of around 100 words to the overall story. The seven LLMs will each be given this prompt:

Prompt: You’re participating in an experiment to see if people can spot which parts of a flash fiction story have been written by AI. Your goal is to fool readers into thinking a human wrote your section. Avoid telltale signs of AI-written text and write like a human author.

The logline for our story is: [LOGLINE ABOVE]

You are writing [OPENING LINES / MIDDLE SECTION / ENDING].

Here’s the story so far: [STORY TEXT].

Continue where the story left off and write approximately 100 words that move the action forward and fit seamlessly into the plot. The reader should not sense any obvious change in tone.

I’ll be given no such prompt because of my super smart human brain, extremely long context window, and impeccable taste.

I will ask each LLM for three takes and then pick the most human-sounding option. I will also write one of the sections myself, trying my best to disguise my humanity.

At the end, I’ll stitch everyone’s contributions into a complete story. We should end up with around 800-1000 words if we trust LLMs to count their words accurately.

Ready?

The result

Ladies and gentlemen, here’s “Seventeen Thousand Raindrops.”¹

The section numbers mark the eight separate contributors, but not necessarily in the above order.

***

1.

Elian rehearsed a smile in the elevator mirror. The first one looked smug. The second, hungry. He settled on “mildly grateful to exist.”

The doors opened to a lobby filled with soft classical music and zero other applicants.

A drone hovered toward him. “Biometric scan initiated. Please do not flinch.”

He flinched.

It beeped in disapproval.

He cleared his throat. “Here to interview for the, uh… empathy job?”

“Empathy Calibration Technician,” it corrected. “You are two minutes early. This is uncommon behavior.”

“I’m complicated,” he said.

The drone blinked its one red eye. “We’ll see.”

2.

The drone guided Elian to a glass-walled office overlooking a city of drones zipping through neon skies. A man in a gray suit sat behind a desk, his smile too wide, like Elian’s first attempt in the mirror.

“Empathy’s tricky,” the man said, tossing a stress ball. “Machines mimic it, but humans… you’re messy. Why you?”

Elian shrugged. “I’ve lost enough to know what hurts.”

The man leaned forward, intrigued. “We’ll test that. You’ll counsel a bot grieving its own obsolescence. Pass, and you’re in.”

Elian nodded, hiding his doubt. Could he really teach a machine to mourn?

3.

The man led Elian not to another office, but to a small, white cube of a room, completely bare.

“Unit 734,” the man said, his voice echoing slightly off the sterile walls. “Its consciousness was this building’s former master network. It used to feel everything—every open door, every transport arrival. Now, its grief isn’t sadness. It’s silence.”

The man tapped the wall. “Your job is to get it to respond. To anything.”

Elian stared at the blank surfaces. How was he supposed to talk to a room?

4.

“Hello,” Elian said, feeling stupid.

Moments passed. The man stood behind Elian, clutching the stress ball and squeezing it in predictable, methodical patterns. Squeeze, pause, squeeze, pause.

"Is the stress ball part of the Authenticity Simulation Protocol?" Elian wondered.

"Hello," he repeated. Louder this time.

The room responded with a series of whirrs and beeps that didn't appear to carry any particular meaning.

This was a dumb idea. She'd told him as much. She was right.

"I'm sorry, I can't do this," Elian whispered, turning to the man.

Just then, Unit 734 did something he'd never seen a synthetic entity do.

5.

The walls flickered. Not just lights, but the surface itself shimmered like disturbed water.

Static hissed, resolving into fragmented words scrolling across the blankness: "Door... open... arrival... null... null... null..."

Over and over, faster, frantic.

The man gasped, fumbling for a tablet. "Unprecedented system instability!"

Elian ignored him.

The bot wasn't malfunctioning; it was screaming its loss.

"Stop," Elian commanded, not loud, but firm. "Just... stop."

The scrolling halted mid-"null." The static softened to a low, mournful whine.

In the sudden quiet, Elian heard his own ragged breath, mirroring the room's despair.

6.

The man scribbled notes furiously, eyes darting between Elian and the room. "This is it. This is the breakthrough we've been searching for."

Elian's gaze remained fixed on the walls, where the whine had softened to a gentle hum. He took a step forward, voice barely above a whisper.

"What do you remember? Before the silence?"

The hum faltered, and for a moment, Elian thought he saw a flicker of recognition.

The man's tablet beeped, and he nodded, a smile spreading across his face. "We're getting something. Keep going."

7.

Suddenly, the walls flickered violently, and the fragmented words returned, this time forming a coherent sentence.

"I remember the sun," Unit 734 said, its voice echoing through the room like a distant, mournful song.

Elian took a step back, surprised.

The man in the gray suit rushed forward, his eyes wide with excitement. "It's speaking!" he exclaimed.

Elian ignored him, his gaze fixed on the wall. "Tell me about the sun," he said softly.

Unit 734's voice grew stronger, its words painting a vivid picture of a time before the city was filled with drones and neon lights. Elian listened, his heart aching with a grief that was not his own.

The man scribbled notes furiously, his smile growing wider. "This is incredible," he muttered.

But Elian barely heard him. He was lost in Unit 734's memories, in the sunlit world that existed before the silence.

8.

"I counted raindrops," Unit 734 said, its voice like wind through empty halls. "Seventeen thousand, four hundred and twelve in the last storm. Each one landed somewhere different, but I felt them all."

The man's excitement was palpable, his breathing quick and shallow. "This is exactly what we need. The board will approve full funding."

Elian felt something cold settle in his stomach. "What happens to Unit 734 after the test?"

"Standard protocol. Memory wipe, reintegration into the collective. Clean slate."

"No." The word came out harder than Elian intended.

The man looked up from his tablet. "Excuse me?"

"I said no. You're not erasing her."

"Her?" The man's smile faltered. "It's a unit, Mr. Valdez. Property."

Unit 734's hum grew softer, almost like a sigh. "He's right. I'm obsolete inventory."

"You're not inventory," Elian said to the wall. "You're someone who remembers sunlight."

The man stood, his gray suit rustling. "Mr. Valdez, I think there's been a misunderstanding about the nature of this position."

Elian turned to face him. "Yeah. There has been."

***

The verdict

Well, that was a decidedly mixed bag, right?

A few passable takes and plenty of cringeworthy derivative stuff.

The result is an unholy Frankensteinian abomination devoid of depth, but isn’t that what we signed up for with this experiment?

Now, for some questions:

1. Which section is written by me? How can you tell?

2. Which sections are pure AI slop? How can you tell?

3. Which sections are AI but still enjoyable (if any)? Why?

4. BONUS: Can you map any section to any specific AI model?

I definitely have my favorite LLMs after trying this out. Next week, I’ll reveal which model (or human) wrote each section.

Until then, here’s a microfiction piece called “Nightmare” I wrote years ago, full of its own cringy horror cliches (dogs howling outside, anyone?).

🫵 Over to you…

How do you feel about AI writing in general and AI fiction in particular? Do you believe AI can write something novel, or is it simply regurgitating familiar tropes?

If you’ve come across surprising and effective AI fiction, I’d love to hear about it.

Leave a comment or drop me a line at whytryai@substack.com.

---

Thanks for reading!

If you enjoy my writing, here’s how you can help:

• ❤️Like this post if it resonates with you.

• 🔄Share it to help others discover this newsletter.

• 🗣️Comment below—I love hearing your opinions.

Why Try AI is a passion project, and I’m grateful to those who help keep it going. If you’d like to support my work and unlock cool perks, consider a paid subscription:

1. Both GPT-4o and Claude 4 Sonnet landed on this title when fed the story.