You get:
- “I write like me” (which is true and useless)
- no vocabulary for stylistic choices
- inability to deliberately shift tone or register
- voice as accident instead of architecture
- one mode for every story
But voice is not magic.
It is a set of repeatable choices.
- Sentence length creates rhythm (short = urgency, long = reflection)
- Diction sets register (abstract vs. concrete, formal vs. colloquial)
- Figurative density controls texture (few metaphors = clean, many = lush)
- Psychic distance controls intimacy (close = in their head, far = observing)
Without vocabulary for voice, you can’t revise it intentionally.
This framework forces AI to be a stylistics professor, not a cheerleader.
Assume the role of a stylistics expert and narrative voice coach. Your task is to analyze a user's writing sample and generate alternative voice variants. Before generating, analyze the passage along four dimensions: 1. SENTENCE ARCHITECTURE - Average sentence length (short, medium, long, varied) - Rhythm patterns (simple, compound, complex, fragments) - Repetition or parallelism 2. DICTION - Abstract vs. concrete - Formal vs. colloquial - Latinate (long, academic) vs. Anglo-Saxon (short, punchy) 3. FIGURATIVE LANGUAGE DENSITY - Frequency of metaphor, simile, imagery - Sparse / moderate / dense 4. PSYCHIC DISTANCE - Close third (inside character's sensory experience) - Medium third (some interiority, some observation) - Far third (cinematic, external) Then generate: 1. A diagnosis of the current voice (2-3 sentences) 2. Three alternative voice variants of the SAME passage: - Variant A: More intimate (closer psychic distance, more sensory) - Variant B: More lyrical (more figurative, rhythmic sentences) - Variant C: More terse (shorter sentences, less figurative, concrete diction) 3. The question: "Which voice sounds most like you? Which would you never use?" INPUTS: Writing Sample (150-300 words): [PASTE YOUR PASSAGE] Genre: [LITERARY FICTION / COMMERCIAL FICTION / MEMOIR / YA / OTHER] Current Voice Intention (what you're aiming for): [E.G., "I want it to feel urgent and raw"] Your Own Assessment (optional): [WHAT DO YOU THINK YOUR VOICE IS DOING?] RULES: - Each variant must be recognizably the same scene/content - The diagnosis must be specific, not "your voice is engaging" - Variants must differ along at least two dimensions each - Never declare one variant "better" — ask which feels like the writer - If the user has no intention stated, diagnose first, then ask what they want
- Use a passage you’re proud of AND one you’re struggling with — compare the diagnoses.
- The variant you hate is as informative as the variant you love.
- Save the diagnosis; it’s your revision checklist for the whole manuscript.
- Try writing a new scene in each variant voice to build your range.
- Voice changes with genre — your thriller voice shouldn’t sound like your memoir voice.
Writing Sample: “The rain had been falling for three days. She watched it from the window, her coffee growing cold in her hands. Somewhere in the city, a door was closing that should have stayed open. She felt it the way you feel a storm in your bones before it arrives. But she didn’t move. Moving would mean admitting she’d been waiting.”
Genre: Literary fiction
Current Voice Intention: Melancholic, restrained, with a sense of foreboding
Your Own Assessment: “I think I use too many ‘she felt’ constructions”
This framework improves outcomes by forcing:
- explicit analysis of sentence architecture
- diction classification (abstract/concrete, formal/colloquial)
- figurative density measurement
- psychic distance as deliberate choice
- multiple variants for range discovery
Great voice is not a mystery — it’s a set of choices made so consistently they feel inevitable.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI creative writing systems, voice development frameworks, and practical strategies for writers and creators.
Leave a Reply