Why Do Most Fiction Prompts Fail?
Type “Help me with my story” into any AI tool and you’ll get garbage back, every time.
You know this because you’ve tried it. You got back something vague, generic, and useless. Maybe a list of tips you already knew. Maybe suggestions so broad they could apply to any story ever written. Then you decided AI doesn’t work for fiction. Wrong conclusion. AI works fine. Your prompts don’t. AI responds to specificity. The more context you provide, the more targeted the output. The more precise your question, the more usable the answer. Fiction writers give AI almost no context and ask impossibly broad questions. Then they blame the tool.
“Help me with my character” gives AI nothing to work with. Which character? What’s their role in the story? What problem are you having? What have you already tried? What kind of help do you want? AI doesn’t know. So it guesses. And guessing produces the generic slop that made you give up on AI in the first place.
Better AI won’t fix this. Learning to ask better questions will.
Watch: 12 Best ChatGPT Prompts for Authors (Don't Write Without These!), The Nerdy Novelist
How Much Context Does a Prompt Need?
Every useful fiction prompt has three layers.
Layer one: Situation. What exists right now? Who’s this character? What’s happened in the story? What’s the current problem? Give AI the facts it needs to understand your specific situation.
Layer two: Constraint. What limitations apply? What genre are you writing? What tone? What have you already tried that didn’t work? What do you want to avoid? Constraints focus the output.
Layer three: Request. What exactly do you want? Not “help.” Specific deliverables. Ten options. An analysis. A list of problems. Concrete output you can evaluate.
A prompt with all three layers produces results. A prompt missing any layer produces mush.
Which brainstorming prompts work for fiction writers?
Bad: “Give me ideas for my villain’s motivation.”
Good: “My villain is a 50-year-old former surgeon who now runs an underground organ harvesting operation.
The story is a crime thriller set in Chicago. I want his motivation to be sympathetic enough that readers understand him without excusing him. Give me ten possible motivations rooted in loss, fear, or twisted idealism. Avoid generic greed or pure sadism.”
The second prompt gives AI a character, a genre, a tone goal, specific constraints, and a concrete deliverable. It’ll produce something usable.
Bad: “How should my romance end?”
Good: “My romance is enemies-to-lovers between a divorce attorney and her opposing counsel.
They’re both cynical about love because of their jobs. The genre is contemporary romance with comedic elements. I need them to end up together, but I want the resolution to address their cynicism instead of ignore it. Give me five possible final scenes that would feel earned instead of convenient.”
Context. Constraint. Concrete request. Every time.
Analysis Prompts That Work
AI is better at analyzing your work than generating new work. That’s where I’d point any fiction writer first, because a machine that critiques your chapter leaves the writing in your hands, where it belongs. Bad: “Is my chapter good?”
Good: “Here’s chapter four of my psychological thriller. Read it and identify: any places where my protagonist’s motivation seems unclear, any moments where pacing drags, and any dialogue that sounds unnatural or expository. Quote specific passages and explain what’s not working.” The specific asks give AI something to look for. Vague asks produce vague answers.
Bad: “What’s wrong with my scene?” Good: “This scene is supposed to build romantic tension between two characters meeting for the first time. Read it and tell me: Does the attraction feel mutual or one-sided? Are there enough moments of physical awareness? Does the dialogue have subtext or is it too on-the-nose? Where does the tension peak and does it peak too early?”
You’ve told AI what the scene should accomplish and asked it to evaluate whether it accomplishes those things. Now the feedback has focus. The Conflict and Tension Handbook covers what tension should look like so you know what to ask AI to find.
Which AI prompts help with character development?
Bad: “Create a character for my fantasy novel.”
Good: “I need a secondary character for my epic fantasy who serves as my protagonist’s mentor.
The protagonist is a young woman discovering she has forbidden magic. I want the mentor to have a complicated past that relates to forbidden magic. Give me five possible backstories for this mentor, each suggesting a different relationship dynamic: one protective, one manipulative, one reluctant, one with a hidden agenda, and one who believes in the protagonist.”
You’ve given the role, the context, and specific variations you want to compare.
Bad: “What’s my character’s wound?”
Good: “My protagonist is a 35-year-old firefighter who can’t maintain relationships. She’s brave at work but emotionally avoidant in her personal life. Her mother was emotionally neglectful. Give me five possible psychological wounds that could explain this pattern, each suggesting different false beliefs she might hold about herself and different ways her arc could resolve.”
Now you’re getting options you can evaluate against your vision. The Deep Character Handbook covers wound psychology so you can recognize which suggestions fit.
Research Prompts That Work
AI is fast at pulling research together for fiction. It’s also confidently wrong often enough that I’d never put a detail from it into a manuscript without checking it somewhere else. A wrong fact about a poison or a castle kitchen is exactly the kind of mistake readers write angry reviews about. (See also The Book of Poisons)
Bad: “Tell me about medieval castles.”
Good: “I’m writing a fantasy with a medieval European setting. My protagonist is a servant in a castle. I need to understand the daily routine of castle servants, the physical layout of servant quarters versus noble areas, how servants would interact with nobility, and specific details about food preparation in a castle kitchen. Focus on sensory details I could use in scenes.”
You’ve told AI what you need the research for and what form you need it in. “Sensory details I could use in scenes” is a filter that shapes the output.
Bad: “How does poison work?”
Good: “In my mystery, the victim is slowly poisoned over three weeks before dying. I need a poison that could be administered in food without detection, would cause symptoms that might be mistaken for natural illness, would be available to someone in 1920s England, and wouldn’t be immediately detected by an autopsy of that era. (See also my review of Deadly Doses) Give me three options with the symptoms and timeline for each.”
Specific constraints. Specific deliverables. Usable output.
How Many Attempts Should a Prompt Take?
Your first prompt rarely produces exactly what you need. Plan for conversation.
Prompt one gets you in the ballpark. Read the output. What’s useful? What’s off-target? Prompt two narrows: “Option three has potential but feels too dark for my tone. Give me three variations on that concept that are more morally ambiguous and less evil.” Prompt three refines: “I like the version where her wound comes from protecting her sister. How might this wound manifest in her relationships with authority figures specifically?”
Treat AI like a brainstorming partner who needs guidance, not a vending machine that should produce finished answers on the first try.
My take
Vague prompts get garbage, and the fix is on your side of the screen. Give every prompt the situation, the constraints and one concrete thing you want back. Ask for three options you can judge. Plan on three rounds, narrowing each time. Treat it like a brainstorming partner who needs direction, and never expect a finished answer from the first try.
Which Prompts Should You Never Use?
“Write this scene for me.” I’d cut this one out of every writer’s vocabulary. AI-written scenes need so much rewriting that you haven’t saved any time, and the scene still won’t sound like you. Brainstorm the scene with AI and write it yourself.
“Tell me if this is good.” AI can’t judge quality. It can analyze for specific criteria. “Good” isn’t specific.
“Make this more interesting.” Meaningless. What does interesting mean? More conflict? More voice? More surprising word choices? Define it or don’t ask.
“Fix my plot hole.” AI can brainstorm solutions to plot holes you’ve identified and explained. It can’t see plot holes you haven’t noticed. That requires human readers.
Every prompt that asks AI to do your creative thinking fails, and it should. The prompts that ask it to support your thinking are the ones worth learning. A writer who hands the machine the decisions ends up with a book nobody decided anything about, and readers feel that from the first page.
| Prompt to avoid | Why it fails | Ask this way |
|---|---|---|
| “Write this scene for me.” | Needs so much rewriting you save no time, and it won’t sound like you | Brainstorm the scene with AI, then write it yourself |
| “Tell me if this is good.” | AI can’t judge quality; “good” isn’t specific | Ask it to analyze against specific criteria |
| “Make this more interesting.” | “Interesting” means nothing until defined | Name what you want: more conflict, more voice, more surprising word choices |
| “Fix my plot hole.” | It can’t see plot holes you haven’t noticed | Identify and explain the hole, then ask for solutions |
Frequently Asked Questions
- Best practices for prompt engineering with the OpenAI API, OpenAI Help Center
- Prompting best practices, Anthropic (Claude Platform Docs)
