By skyforbes FYI, this was all typed up by hand, it’s not an AI generated post. Minus the examples at the end. Please forgive any formatting issues; i’m posting this on my phone and sometimes that messes things up. TLDR: Gemini is vastly superior at creative writing, and I personally believe anybody who says otherwise is wrong.
Anyway. I’m an author who likes to use AI for certain steps in the writing process. Pretty much everything but the writing itself. (See my profile posts for more detail). I also mess around with creating a lot of 100% AI stories just to read myself. Think in the 20k-40k word range. For example, I had this wacky idea this morning of what would happen if, in the 1969 moon landing, Neil Armstrong saw something walking on the moon. That something turned out to be a man who gave him a paper stating that aliens would arrive in 2003 and the government needed to prepare, and then vanished in thin air.
Pretty random, right? Yeah. But it’s enjoyable to turn story ideas like that into an actual story just for funsies.
Now let’s get down to the point of this post. In my opinion, which seems to be an unpopular opinion, Gemini is far and away superior for creative writing. It’s not even close. ChatGPT just can’t help but use all kinds of metaphors, performative sentences, and dramatic, cliche lines. No matter what you tell it, no matter how much you try to tell it not to do something, it writes the opposite of “show, don’t tell.” Sure, it might have a better memory, and it might be able to look up facts in the real world a little bit better… but it’s simply awful. Unless I’m just missing sometime important, which I doubt, because I’ve tried everything. Yes, I know all about custom instructions, etc.
With ChatGPT, it doesn’t really matter how many times I go back through and say, “hey don’t do this, don’t do that, here’s an example of my writing”… it just can’t help but pump out all kinds of dramatic bullshit. With Gemini, it’s better from the get-go, and I can keep feeding my edits back to it and it actually learns how to write better/ imitate. ChatGPT really struggles with that.
Let’s go through some examples. (I used the exact same prompts for both models). To start with, I fed both Gemini and ChatGPT my 6000 word long “starter prompt.” That prompt has taken me a very long time to get where I want it to be, and gives very specific examples of my writing style, etc. I won’t copy that here obviously, but it’s not vague at all. I even made both chats for the story in a new project, and had the starter prompt put into the custom instructions, telling it to make sure it checks that the outputs are correct. That seems to keep it on track. Well, Gemini, that is.
And then, after the starter prompt, I made Gemini and ChatGPT give me a very detailed outline on what it might look like from the POV of Mission Control in that particular example of a story. ChatGPT is definitely better for stuff like that. It took a lot longer, sure, but the outline had more real-world detail, so I used the ChatGPT outline for both stories. However, when it came to the writing itself…
Here’s two snippets of the story from ChatGPT:
Dr. Vance didn’t like the viewing room.
It was too clean, too new, too far from where the work actually happened. The glass made it feel like a museum exhibit, like the people on the other side were a diorama you could point at and explain to visitors. He’d fought for his chair at this hour anyway, because the Administrator was supposed to be seen.
It wasn’t just about doing the job; you had to be seen doing it.
The room was crowded with the right kind of quiet. Not silence, not reverence, just that professional hush you got when everyone in the building was listening to the same handful of voices.
A few feet beyond the glass, Mission Control sat in its tiers like a small arena. Consoles with clipped labels, headsets, loose papers stacked in neat nervous piles. Coffee cups that had been filled and forgotten. Ashtrays that were technically there but mostly unused tonight. The air smelled like electronics and warm plastic and the faint sour edge of sweat, the smell you couldn’t scrub out of a place that had been worked too hard for too long.
Vance stayed where he was. The viewing room had a way of making him feel useless, and he didn’t like that either. He preferred hallways and phone calls and decisions that had to be made in real time. Here, he could only watch.
On the public feed, the picture steadied again. The ladder shape stayed in frame longer now, and the smear around it looked more like ground instead of noise.
It still wasn’t good. It didn’t have to be. Everyone already knew what they were looking at.
CLIP FROM LATER ON:
One of the staffers in the viewing room spoke softly, not to Vance directly, but close enough that Vance heard it. “We’re live to the networks.”
Vance didn’t nod. He didn’t need to. The world would see whatever this camera could show, and they’d fill in the rest in their heads.
Armstrong’s voice came through again, measured and plain. He was narrating what he saw the way he’d been trained to do, because the record mattered.
“I’m at the foot of the ladder, and the LM footpads are only depressed in the surface about one or two inches.” -END
I won’t even get into all the issues I have with that or why it’s so terrible and cliche. That would be another 10 paragraphs in itself. But for a few examples: every fucking line is an example of told, not shown, even though my starter prompt is extremely clear about not doing that. Not to mention that I told it to never use “It’s not X, it’s Y,” and it used that. And also, even though I explicitly told it to vary the sentence structures, the vast majority of all those sentences have a single comma in them. That pulls the reader out of the story. Honestly, the only positive thing I can say is that it now finally doesn’t use a buttload of em-dashes.
Now for Gemini:
Dr. Harold Vance hated the viewing room. It was a fishbowl, separated from the real work by a thick pane of soundproof glass. Down in the trench, the floor-wax smelling air was recycled and cold. Up here it smelled like stale cigarette smoke. He stood near the railing, hands in his pockets, watching the rows of consoles below.
The Mission Operations Control Room was a sea of white shirts and skinny black ties. It looked organized and calm from his viewpoint, a grid of men staring at green cathode-ray tubes, but Vance knew better. He knew that every man down there was holding his breath as tightly as he was.
"Networks are live," the PR aide whispered, leaning in so close his breath fogged Vance’s wire-framed glasses. "Thirty seconds to the broadcast cut-in."
Vance nodded after stepping back. On the far wall, massive projection screens dominated the room. The center screen displayed the ground track as a sine wave creeping over the gray map of the lunar surface. To the left was the video feed. It was garbage. A grainy, high-contrast smear of gray and black that kept rolling with static.
"Hey fellers, can y’all clean that up?" a random senator drawled from the back of the room, jabbing his cigar toward the screen.
CLIP FROM LATER ON:
The senator's cigar glowed briefly, then dimmed again as he drew on it. Vance pretended not to notice. The fact that cigars weren't supposed to be lit today was a fight for a later time. Hopefully a battle someone else would wade into.
Armstrong's voice came through. "I'm at the foot of the ladder. The footpads are only depressed in the surface about one or two inches."
"Copy that," CAPCOM replied.
The video wobbled as if the camera couldn't decide what it wanted to show; the ladder in frame, the black sky, or the gray surface that looked like powdered concrete. It kept flickering at the edges with snowlike static before returning.
"Y’all sure we can’t get that any clearer?" the senator huffed again.
Vance didn't respond this time. Nor did anyone else, as static mixed with breathing became the primary sound. "Okay," Armstrong stated. "Preparing to step off the LM." -END
Now, does that still have issues? Of course. And I certainly wouldn’t consider it on par with my actual writing. But it’s better in virtually every single aspect. Gemini added a lot more dialogue and interesting little tidbits without so much stupid drama and falsified tenseness. I didn’t even tell it to add in the senator, it did that itself, and even had him use a bit of regional dialect to spice things up.
This post is already crazy long, so I won’t go into more detail. Feel free to leave a comment if you feel differently… at this point, judging by all the “ChatGPT is SO MUCH BETTER at creative writing” I’m just hoping I’m doing something wrong. But Gemini, even if Gemini 3 isn’t really any better at creative writing than 2.5, seems to me like the clear winner.