If you’ve ever asked an AI to help you develop a character, write a scene, or untangle a tricky moment in your draft — and gotten back something that sounds like every other novel you’ve ever read — you’re not alone. And the AI isn’t broken.
The problem is almost always the prompt.
The uncomfortable truth is most authors are asking AI questions when they should be giving it direction. And character-driven fiction punishes that mistake harder than any other kind of writing, because the moment the AI defaults to its training data’s idea of “a character,” you get a cardboard cutout in a trench coat.
Let’s talk about why that happens — and a few foundational shifts you can make right now to start getting prompts that actually serve your fiction.
You’re Not Asking AI a Question. You’re Directing It Toward an Outcome.
This is the single biggest mindset shift, and it changes everything downstream.
When most writers prompt an AI, the structure looks like this:
“Can you help me develop my main character?”
Or:
“Write a tense conversation between these two people.”
The AI happily complies. It gives you something. And the something is usually fine. Usable. Forgettable.
That’s because you didn’t tell the AI what “done” looks like. You asked it to perform a vague task, and it performed a vague task. Garbage in, generic out.
Compare that to:
“I need a character profile that gives me one specific psychological contradiction this character is living inside — not a trait, not a backstory, an irreconcilable internal conflict. Output: one sentence stating the contradiction, plus three observable behaviors that hint at it without naming it.”
Same request, fundamentally different prompt. The second one tells the AI exactly what “done” means. It defines the shape of the output. It tells the AI what not to do (don’t give me traits, don’t give me backstory). It gives the AI a target to aim at.
This is the foundational shift: define done before you prompt. Know what you want on the page when the AI is finished. If you can’t describe what you want in concrete terms, the AI definitely can’t produce it.
Context Beats Cleverness Every Time
The second mistake authors make is starving their prompts.
You’ll see writers craft a beautifully worded prompt — clever, specific, full of personality — and then give the AI almost nothing to work with. No character context. No scene context. No story context. Just the prompt and a hope and a prayer. (I’m sure you’ve been there.)
Meanwhile, the AI is sitting there with its training data — which is millions of generic novels — and absolutely no information about your book. So it pattern-matches to the most likely shape of what you might be writing, which is, almost by definition, the most clichéd version possible.
Give the AI what it needs to do the task. Not everything you know about your book — only what’s relevant for this specific output. If you’re prompting for a scene, give the AI:
- Who’s in it and what they want from each other
- Where they are and why that matters
- What just happened before this moment
- What the scene’s job is in the larger story
- What the prose register should sound like
That’s it. That’s the difference between “the AI gave me something generic” and “the AI gave me something I can actually use.”
You are not being clever by withholding context. You are not being efficient. You are starving the model and then blaming it for the crappy meal it just served. (Be nicer.)
The Prompt Killer Almost No One Catches: Contradictions
Most prompts that fail aren’t failing because they’re too short or too vague. They’re failing because they contain contradictions — instructions that cannot both be satisfied at the same time. And the AI, being a probability machine, just averages between them and gives you something that satisfies neither.
Real examples I see authors write all the time:
- Tone clash: “Make it professional but also fun and a little snarky.”
- Length clash: “Give me a quick one-paragraph summary, and include all the key details and examples.”
- Format clash: “Give me a bulleted checklist, but write it in a story style.”
- Goal clash: “Give me three options, but also tell me the best one, and don’t ask any follow-up questions.”
- Scope clash: “Keep this high-level, but I need step-by-step instructions I can follow today.”
Look at those. Each one sounds fine. Each one is asking for two genuinely incompatible things. And every time, the AI tries to split the difference and produces something muddy.
The fix is straightforward, even if it requires honesty:
- Decide what matters most. What’s the non-negotiable? Tone? Length? Format?
- Demote the rest from constraints (must-have) to preferences (nice-to-have).
- Add a priority rule to your prompt: “If anything conflicts, prioritize clarity over style.” Or “If you have to choose, prioritize accuracy over brevity.”
That one line — the priority rule — fixes more bad outputs than any other prompting trick I know. Try it this week. Add a priority instruction to your next three prompts and watch what happens.
Why This Matters Ten Times More for Character-Driven Fiction
Genre fiction has scaffolding. Plot beats. Tropes. Expected shapes. When the AI defaults to convention, you can usually rescue the output because the convention is part of what you wanted.
Character-driven fiction is the opposite. Its entire job is to escape conventional shapes. Its entire job is to find the specific, the strange, the contradictory, the human. The moment the AI defaults to a recognizable template — “the damaged loner,” “the reluctant mentor,” “the redemption arc” — your character dies on the page.
Which means everything we just talked about — defining done, providing context, eliminating contradictions — isn’t a nice-to-have. It’s the difference between a character who feels like a real person and one who feels like a Mad Lib of fiction tropes.
That’s just the foundation. Once you know how to direct AI rather than ask it questions, the next layer is figuring out how to direct it toward the kinds of character work that resist defaulting in the first place — work built on contradiction, wound, and the specific instead of the general.
This is where things get really interesting. But it’s also where most authors stop, because the foundational moves alone produce better output, but they don’t produce unforgettable output.
Where to Go From Here
If you only do three things this week:
- Define “done” before you prompt. Write down what the output should look like before you write the prompt itself.
- Front-load context. Tell the AI what it needs to know about your story to do the task — not everything, just what’s relevant.
- Audit for contradictions. Read your next prompt and ask yourself: can these two instructions both be true at the same time? If not, pick one and demote the other.
Those three moves alone will lift the quality of every AI output you generate. Promise.
If you want to know how to direct AI toward genuinely lateral, character-driven fiction; how to build prompting systems instead of one-off prompts; how to use techniques like assumption excavation, provocative operations, and random entry to keep AI from defaulting into the same five character archetypes; how to develop characters from contradiction and wound rather than role and trait; and how to take a raw idea all the way through to a drafted chapter using a repeatable, reusable workflow, that’s exactly what the Prompting for Character Driven Fiction standalone class at Future Fiction Academy is built for.
It’s the full system. Not just the foundational shifts, but the layered prompting workflow that takes you from “I have a vague idea” to “I have a story bible, an outline, and chapters that don’t sound like every other AI-assisted novel out there.” Every prompt is included. Every step is walked through. Every audit is built in.
If you’ve been getting flat character output and you’re tired of it — this is the class that fixes it at the root.
Your AI Isn’t Bad at Character-Driven Fiction
It’s just been trained on a billions of words and, without direction, it will default to the average of all of them.
Your job isn’t to ask better questions. It’s to give better direction.
Define done. Provide context. Kill contradictions. Then keep going from there.
Your characters will thank you.
