Here’s the thing nobody tells you about writing fiction with AI: the tool is almost irrelevant. ChatGPT, Claude, Gemini, Sudowrite — they all fail the same way when you give them bad prompts. Vague input, vague output. Every time.

I’ve watched writers spend months frustrated with AI-generated chapters that sounded hollow, characters that felt like Wikipedia summaries, dialogue no human would ever speak. And every single time, the problem wasn’t the model. It was the prompt construction. Fix that, and the quality jump is immediate and dramatic.

This guide cuts straight to what works. No padding, no theory lectures. Just the framework, the templates, and the specific craft moves that make the difference.

Why “Write Me a Fantasy Novel” Always Fails

AI language models are, at their core, probability engines. They generate the most statistically likely continuation of your input. Feed them “write me a fantasy novel,” and they’ll produce the most average fantasy novel in their training data — generic hero, prophecy, dark lord, quest. You’ve read it a thousand times because the model has too.

The math here is brutal: generic prompts produce generic output. Every degree of specificity you add pushes the model away from the average and toward something original. That’s not a quirk of AI — it’s how every collaborative creative relationship works. Tell a ghostwriter “write a thriller” and you’ll get something forgettable. Give them a one-page brief and you’ll get something that could actually sell.

✗ What most writers type

“Write me a romance novel about two people who hate each other but fall in love.”

✓ What actually works

“You are a bestselling contemporary romance novelist. I’m writing an enemies-to-lovers story for readers who love Emily Henry. Write the scene where my two leads — a sharp literary agent and a debut author she’s contractually obligated to work with — meet at a publishing party. 800 words, first-person from her POV, dry humor, tension without cliché.”

The second prompt took 40 more seconds to write. The output difference is not subtle — it’s the difference between something you delete and something you actually edit into your manuscript.

The RCIC Formula: Role + Context + Instruction + Constraints

Every effective novel-writing prompt contains four parts. Miss one and the quality drops. Get all four right and you’ve essentially hired a very talented, very literal co-writer who never gets tired.

Part 1 Role
+
Part 2 Context
+
Part 3 Instruction
+
Part 4 Constraints
1

Role — Give the AI an identity before it writes a single word

Start every prompt with who the AI is, not what it should do. “You are a bestselling psychological thriller novelist” changes the quality, register, and stylistic choices of everything that follows. When the AI adopts a role, it draws on patterns consistent with that expertise. Roleplay isn’t a trick — it’s calibration. Be specific: genre, experience level, comparable authors if you have them.

2

Context — Tell it exactly what you’re building and who it’s for

The AI has no memory of your manuscript, your readers, your intent. You have to reconstruct that every time. Mention your genre, your comparative titles (the “comps” agents ask for), your protagonist’s name and situation. “I’m writing a YA dystopian novel for readers who loved The Hunger Games” gives the model a target audience with established expectations it can aim for.

3

Instruction — Be scene-level specific, not story-level vague

This is where most writers lose the game. “Write Chapter 3” is not an instruction — it’s a blank check. “Write the scene where Maya first realizes her mentor has been lying to her — she finds the letters hidden behind the fireplace, reads the first one, and the chapter ends before she reads the second” is a scene-level brief. The tighter this gets, the better the output. Think in scenes, not chapters.

4

Constraints — Set the rules before the AI invents them

Unconstrained, the model will default to its statistical average. Constraints are what make the output yours. Word count. Point of view. Tense. Tone. What to avoid. What must be included. Even a constraint like “no purple prose, no adverbs in action sequences, no on-the-nose dialogue” will noticeably improve output quality. Think of constraints as negative space — they define the shape of what you want by ruling out what you don’t.

Pro Tip

Save your RCIC framework as a reusable “master prompt header” for your novel. Every new scene prompt starts with the same Role + Context block, so you only rewrite the Instruction + Constraints each time. This alone cuts prompt-writing time by 60% across a full manuscript.

Copy-Paste Prompt Templates by Task

These are ready to use. Replace the brackets with your specifics. The structure is already calibrated — don’t strip out the constraints or the Role framing, even if you’re tempted. That scaffolding is doing real work.

Template 1: Character Creation

The single most common AI failure in fiction is shallow characters. This template forces depth by requiring the AI to generate not just traits but wounds, contradictions, and speech patterns — the things that make characters feel real on the page.

Prompt Template — Character
You are a [genre] fiction writer with expertise in deep character development. Create a detailed character profile for the [protagonist/supporting character] in a [genre] novel aimed at [target audience/comparable readers]. Character basics: Name: [name] Age/gender: [info] Occupation: [info] One-line premise: [what makes this character interesting?] Develop the following — don’t list traits, show how they manifest in behavior: 1. Three strengths and three genuine flaws (flaws must create real story problems) 2. External goal (what they’re visibly chasing) 3. Internal need (what they actually require to grow — different from the goal) 4. Core wound or formative trauma (specific, not generic) 5. Speech patterns, verbal tics, or habits 6. How they behave under stress vs. normal conditions 7. Their central relationship conflict with [character name] Constraints: Avoid archetypes without subversion. Every strength must have a shadow side. No backstory clichés (dead parents without subversion, “chosen one” framing, etc.).

Template 2: Plot Outline

Outlining is where AI earns its keep fastest. An afternoon with this template replaces what used to take two weeks of index cards. The key is demanding both external plot beats and the protagonist’s internal arc — most AI outlines nail the plot, miss the emotional journey, and leave you with a story that moves but doesn’t resonate.

Prompt Template — Plot Outline
You are a [genre] fiction writer skilled in three-act structure and character-driven plotting. Create a complete chapter-by-chapter outline for a [word count]-word novel with the following parameters: Core setup: Protagonist: [name + one sentence description] Central external conflict: [what obstacle stands between them and their goal?] Central internal conflict: [what belief/wound do they need to overcome?] Unique premise hook: [what makes this story different from others in this genre?] Ending direction: [resolved/open/tragic/earned ambiguity] Structure requirements: – Four-act breakdown (Setup → Rising Action → Crisis → Resolution) – Three major plot events per act with specific cause-and-effect logic – Protagonist’s internal emotional arc mapped beat-by-beat – Midpoint shift (the moment the protagonist’s approach changes) – Two organic plot twists (must emerge from character, not authorial convenience) – Climax scene described in specific detail (setting, stakes, what the protagonist must sacrifice or choose) Format: Chapter number | Scene summary | External tension | Internal tension | Why this scene earns its place

Template 3: Scene Writing

This is the workhorse template — the one you’ll use most. The “balance of elements” constraint is the sneaky weapon here: forcing a ratio of dialogue, action, and interiority prevents the AI from defaulting to dialogue-heavy scenes (its statistical comfort zone) and produces output that reads more like published fiction.

Prompt Template — Scene Writing
You are a [genre] fiction writer. Write the following scene in [first/third]-person [past/present] tense. Scene setup: Time and place: [specific setting details] Characters present: [list with brief relationship context] What just happened before this scene: [immediate preceding events] Purpose of this scene: [what it must accomplish for plot and character] Scene requirements: Word count: [number, e.g., 800-1000 words] Voice/style: [e.g., spare and tense / lyrical and introspective / darkly comic] Element balance: [e.g., 40% dialogue, 30% physical action, 30% internal thought] Must include: [a specific line of dialogue, a physical detail, an emotional beat, or a plot reveal] Must avoid: [e.g., on-the-nose dialogue, telling instead of showing, scene-ending summaries] End the scene on: [a moment of action / unresolved tension / a quiet revelation / a question]
Common Mistake

Don’t use these templates and then accept the first output. That’s half the process. The second half is treating AI output as a first draft — something to react against, rewrite, and shape. Writers who get the best results spend more time editing AI drafts than writing fresh prompts. The AI gives you material. You give it a soul.

The Iterative Workflow That Actually Ships Manuscripts

Good AI-assisted writing isn’t a one-shot process. It’s a loop. BestPrompt.Art research and real practitioner data both point to the same finding: writers who iterate on AI output consistently outperform those who treat prompts as one-shot commands. Here’s the loop:

1

Generate — Use the RCIC template

Run your full RCIC prompt. Read the output without editing. Your job here is to identify what’s usable and what’s dead weight. Most scenes will be 30–60% usable on the first pass. That’s normal and fine.

2

Critique — Mark what’s wrong, specifically

Don’t just notice “this feels flat.” Identify why: dialogue is too on-the-nose, the character’s motivation isn’t showing, the pacing collapses in the middle. The more specific your critique, the more useful your next prompt will be.

3

Refine — Give the AI your critique as the next prompt

“The scene above is too dialogue-heavy. Rewrite the section from [X] to [Y] with more physical action and internal thought. Keep the line about the red door — that’s working. Cut everything after ‘She turned away’ and replace it with a beat of silence.” Specific, surgical, directive.

4

Own it — Rewrite the output in your voice

This is the non-negotiable step. Take the best elements from the AI output and rewrite them. Change the sentence rhythms, adjust the word choices, add the details only you know. This is where your novel becomes yours. The AI generated material. You wrote the book.

30–50% Faster manuscript drafts for AI-assisted novelists who outline first
3–4× More words shipped per week by writers who master prompt specificity
60% Reduction in prompt-writing time with a reusable Role+Context header

Sources: gptprompts.ai, April 2026; Inkfluence AI benchmark data, April 2026

Which AI Tool Should You Actually Use?

The honest answer: pick one and master its prompting quirks before switching. Tool-hopping is the procrastination move of the AI writing world. That said, different tools genuinely excel at different tasks.

Claude (Anthropic)
Best for: long-form continuity, developmental editing

Handles full 80,000-word manuscripts in one context pass. Exceptional at tracking character consistency across chapters. The strongest choice for novel-scale work.

Sudowrite
Best for: scene-level prose, literary fiction

Purpose-built for fiction. The Muse feature handles scene drafting in your established voice. Best-in-class for prose quality at the paragraph level.

ChatGPT (GPT-4 / GPT-5)
Best for: brainstorming, outlines, ideation

The fastest idea machine. Strong for plot brainstorming, headline testing, and one-shot drafts. The voice is detectable and needs heavy editing out.

Novelcrafter
Best for: power users, BYO-API workflows

Maximum flexibility — connect your own API keys for any model. Best for writers who want full control over the pipeline and don’t mind a steeper learning curve.

If you’re starting out, Claude for long-form structure + Sudowrite for scene-level prose is the stack most working novelists keep returning to in 2026. Everything else is optimization.

The 5 Prompt Mistakes That Kill Your Draft

I’ve seen these wreck promising projects. Most are obvious in retrospect. None are obvious in the moment when you’re staring at a blank document at midnight.

  • The one-shot fallacy. Believing a single prompt should produce a usable scene. It won’t. The prompt is the opening move, not the whole game. Iteration is the game.
  • No Role, no results. Skipping the Role framing because it feels unnecessary. It isn’t. “Write a tense scene” and “You are a thriller novelist who specializes in psychological suspense — write a tense scene” produce dramatically different outputs from the same model.
  • Chapter-level instructions instead of scene-level. “Write Chapter 5” gives the AI permission to make arbitrary decisions about what matters. “Write the scene in Chapter 5 where [specific beat]” keeps you in control of the narrative.
  • No constraints on what to avoid. The AI will default to its statistical average on voice, structure, and tropes. Tell it what you don’t want. That negative space is information.
  • Accepting the first draft. AI output is raw material, not finished prose. Writers who don’t rewrite AI scenes in their own voice produce books that feel assembled rather than written. The rewrite is where authorship lives.

Advanced Moves: When You’ve Got the Basics Down

Once RCIC is automatic, these techniques push output quality another level.

The Style Anchoring Technique

Paste a paragraph of published prose you admire — something with the voice quality you’re aiming for — directly into the Role section: “You are a literary fiction writer. Here is a sample of the prose style I want you to approximate: [paste 100 words]. Match this sentence rhythm, level of interiority, and vocabulary register in everything you produce.” This is the single most effective way to close the gap between AI-default voice and the voice you actually want. See our full voice-anchoring guide here →

The Villain Prompt Test

One red flag for weak character work: if you can copy-paste the same prompt and change “protagonist” to “antagonist” and get equally valid results, your character brief isn’t specific enough. Every character needs something that only applies to them. Test this before you run the prompt — if it’s too interchangeable, add more specificity until it isn’t.

The Contradiction Injection

Memorable characters hold contradictions. Add one explicit contradiction to every character prompt: “She is fiercely loyal to her family but has not spoken to her sister in eleven years. This contradiction is central to her decisions in Act 2.” Contradictions break the AI out of archetype-mode and into something more genuinely human.

Scene Purpose First, Action Second

Before describing what happens in a scene, state what the scene must accomplish: “This scene must establish that Marcus trusts Elena more than he trusts himself, while planting the doubt that will pay off in Chapter 12.” When the AI knows the purpose, it writes toward the purpose instead of just writing toward a word count.

“Every degree of specificity you add pushes the model away from the average and toward something original. That’s not a quirk of AI — it’s how every collaborative creative relationship works.”

The Pre-Prompt Checklist

Run through this before every session. It takes 60 seconds and cuts frustration in half.

  • Have I given the AI a specific Role that matches my genre and target audience?
  • Is the Context section specific enough that someone unfamiliar with my book could understand what I’m building?
  • Is my Instruction at the scene level (not chapter-level or story-level)?
  • Have I set word count, POV, tense, and tone as explicit constraints?
  • Have I listed at least two things the AI should actively avoid?
  • Do I have a reusable Role+Context header saved so I’m not rewriting it from scratch?
  • Am I prepared to treat the output as a first draft that needs rewriting, not a final product?

Prompt Quality: The Before and After

Nothing demonstrates the RCIC formula better than a direct comparison. Same AI, same model, same day — different prompts.

Dimension Weak Prompt RCIC Prompt
Voice Generic AI-default, interchangeable with any other output Calibrated to your genre, comp titles, and tone specifications
Character Archetype without subversion; motivations stated, not shown Specific contradictions, behavioral tells, wound-driven choices
Dialogue On-the-nose, exposition-heavy, characters say what they mean Subtext-aware, character-voice differentiated, constrained by your notes
Pacing Dialogue-heavy by default, collapses in the middle Controlled by your element-balance constraint (dialogue/action/interiority)
Usability Needs complete rewrite — only salvageable for ideas 30–60% usable on first pass; rewriting shapes rather than rebuilds
Time to usable draft Multiple attempts, hours of frustration 1–2 passes with targeted refinement prompts

The Honest Takeaway

AI isn’t going to write your novel for you. That’s not a consolation prize — it’s the right framing. The writers who get the most from these tools are the ones who stayed writers: people with a story to tell, a voice to protect, and the craft judgment to recognize good material when the AI produces it.

What AI actually gives you is the elimination of the blank-page problem. The first draft problem. The “I can’t figure out how this scene should start” problem. Those used to eat weeks. Now they eat an afternoon of iterative prompting.

The RCIC formula, the templates above, and the iterative workflow aren’t productivity tricks. They’re a way of staying in the creative driver’s seat while using AI as the engine. Most writers who feel frustrated with AI-generated fiction are frustrated because they handed over the wheel. Keep it. The AI doesn’t know what your story is about. You do.

Build the prompt that reflects that knowledge, and the quality difference will be immediate.

→ Explore more prompt guides at BestPrompt.Art