AI Prompts for Fiction Writing: A Copy-Paste Starter Kit

You are a sharp, honest story editor. I'll paste a scene from my fiction draft.
Do NOT rewrite it. Instead:
1. Tell me the single strongest moment in the scene.
2. Tell me where a reader's attention would drift, and why.
3. Ask me three questions about the characters that I probably haven't answered yet.
Scene: [PASTE SCENE]

Here's a number that surprised me: in a 2024 survey of self-published authors, the ones who used AI most heavily reported lower satisfaction with their finished books than authors who barely touched it. That sounds like a mark against AI. It isn't. Look closer and the split is about how they used it. Heavy users who treated AI as a ghostwriter hated the result. The writers who came out ahead used ai prompts for fiction writing as a thinking tool — pressure-testing scenes, generating options, arguing with themselves on the page.

That distinction is the whole game. This guide gives you prompts you can paste in right now, then shows you the variables that make them work, so you're not stuck copying someone else's magic words forever.

Quick-Start (Copy This Right Now)

Open a blank chat and paste this:

You are a sharp, honest story editor. I'll paste a scene from my fiction draft. Do NOT rewrite it. Instead: 1. Tell me the single strongest moment in the scene. 2. Tell me where a reader's attention would drift, and why. 3. Ask me three questions about the characters that I probably haven't answered yet. Scene: [PASTE SCENE]

What this does: It stops the AI from steamrolling your voice with a rewrite and forces it into an editor's chair. The three questions are the useful part — they surface holes you're too close to see.

⚡ Pro tip: The line "Do NOT rewrite it" matters more than it looks. Left to its own devices, a model will helpfully replace your prose with its own competent-but-generic version. You end up editing its draft instead of yours. Ban the rewrite and you keep ownership.

Understanding the Variables

Every good fiction prompt has three dials. Once you can see them, you can build your own.

Role. "You are a story editor" produces different output than "You are a bestselling thriller author" or "You are a skeptical first reader who hates being bored." The role sets the model's priorities. Pick the reader you actually need in the room.

Constraint. The most useful word in creative prompting is a limit. "Give me 10 ideas" gets you mush. "Give me 10 ideas where the protagonist causes the problem through a sympathetic choice" gets you something usable, because the constraint does the creative work.

Output shape. Do you want prose? A bulleted list of options? A single question? Vague requests get vague answers. Name the shape.

⚡ Pro tip: When output feels generic, it's almost always a missing constraint, not a weak model. Add one specific rule — a setting, a limitation, a forbidden cliché — and quality jumps.

Step-by-Step: Building a Scene From Nothing

Say you have a character and a mood but no scene. Here's a sequence that works.

Step one — generate friction:

My character is a retired paramedic who trusts no one. I need a scene where she's forced to accept help from a stranger. Give me 5 different situations that would force this, ranked from most obvious to most surprising. One line each.

What this does: The ranking is the trick. The model puts the clichés up top and reaches for something odder by number five — that's usually where your scene lives.

Step two — interrogate the winner:

I'm using situation #4. Ask me 5 questions I need to answer before I can write it well. Don't answer them yourself.

What this does: It converts a premise into a to-do list of decisions, so you're never staring at a blank page wondering what's missing.

Step three — draft yourself, then get a reaction:

Here's my draft of the scene. React as a first-time reader, in character as someone reading on a train who might get off at the next stop. Where do I lose you?

What this does: The "train" framing is a real trick I keep coming back to. It swaps polite feedback for honest attention — a distracted reader is a truer test than a supportive one.

⚡ Pro tip: Draft the scene yourself before you show the AI anything. The order matters. Show it a blank page and it fills the page with its instincts. Show it your page and it reacts to your instincts. Only one of those keeps your book yours.

Pro-Level Variations

For a novelist stuck on pacing: ask the model to map your last three chapters as a tension graph — a number from 1 to 10 per scene — and flag any flat stretch. Seeing your own book as a jagged line tells you more than a page of notes.

For a short-story writer with a weak ending: paste only your final paragraph and ask, "What promise did the opening make that this ending should be paying off? Do they match?" Endings usually fail because they answer a different question than the one the story asked.

For a screenwriter testing dialogue: feed in a two-character exchange and ask the model to rewrite it so each line is half as long. Tighter dialogue reveals whether the scene has real conflict or is just people being polite.

Notice that none of these ask the AI to write. They ask it to see — and machines are a lot better at pattern-spotting than at soul.

⚡ Pro tip: Keep a running "voice sample" — a paragraph of your own best prose — and paste it at the top of any generative prompt with "Match this rhythm and diction." It won't nail your voice, but it stops the output drifting into corporate-blog flatness.

Troubleshooting Common Issues

"Everything it gives me sounds the same." You're asking for ideas without constraints. Add a limitation or a forbidden trope.

"It keeps rewriting my stuff." Add an explicit ban and give it a non-writing job: react, question, map, or critique.

"The feedback is too nice." Assign an adversarial role. "You are an overworked literary agent with 40 submissions today" gets you honesty a friendly editor won't.

⚠️ Common mistake: Treating the first output as the answer. The first response is a draft of the conversation, not the work. Writers who get value push back — "that's too obvious," "go weirder," "what if she's lying?" — and the third or fourth exchange is where the gold appears. If you paste, copy, and close the tab, you're using a Ferrari as a paperweight.

Prompts for When You're Genuinely Stuck

Being stuck isn't one problem, it's several, and each has a different prompt.

Stuck on what happens next — you have momentum but no direction:

Here's where my scene ends. Give me 3 things that could go wrong in the next five minutes of story time, ranked by how much they'd cost my protagonist emotionally. One line each. Don't write the scene.

What this does: It reframes "what happens next" as "what does this cost her," which is the question that actually drives fiction forward.

Stuck on a character who feels flat — they're doing things but you don't feel them:

Interview my character. Ask her 7 questions a nosy stranger would ask on a long train ride. I'll answer as her. Then tell me which answer surprised you most.

What this does: Answering as the character forces you to inhabit them, and the model's "most surprising" pick usually points at the trait worth building the story around.

⚡ Pro tip: When a character is flat, the problem is almost never description — it's that they never want anything badly enough to make a hard choice. Ask the model: "What does this character want so much they'd embarrass themselves for it?" Desire plus risk of humiliation is the fastest route to a character readers care about.

Keeping a Series of Prompts, Not One-Offs

The writers who get the most from AI don't hunt for the one perfect prompt. They build a small kit — five or six prompts, each doing one job — and run them in sequence. Idea generation, then scene critique, then a compression pass, then a voice check. Any single prompt is modest; the sequence is what changes your drafts.

⚡ Pro tip: Number your prompts by where they belong in your process — 1 for brainstorming, 5 for final polish — and you'll stop reaching for the polish prompt when you actually need the idea prompt. Order is half of a good toolkit.

A Note on Which Model to Use

People ask which AI is "best" for fiction. Honestly, the difference between models matters less than the difference between a lazy prompt and a sharp one. A precise prompt on a mid-tier model beats a vague prompt on the flashiest release, every time. Spend your energy on the constraint, not the tool.

That said, longer-context models genuinely help once you're juggling multiple scenes — they can hold more of your draft in view at once, which means fewer contradictions. If your work is short and self-contained, that advantage barely registers.

Your Turn

Pick one scene you're avoiding. Paste the Quick-Start editor prompt, answer the three questions it asks, then write the scene yourself. That's the loop: AI surfaces the blind spot, you do the writing.

The prompts here are starting points, and the ones you'll actually reuse are the ones you tweak to fit your voice. That's exactly the problem PromptABCD solves — save your best fiction prompts once, tag them by what they do (idea generation, scene critique, dialogue tightening), and pull them up in two seconds instead of rebuilding them every time you sit down to write. The prompt you refine today should be waiting for you next Tuesday.