(So It Doesn’t Feel Bloodless)
Most AI stories fail before the first sentence gets written.
Not because the AI can’t write. It can mimic prose well enough. The problem’s simpler: people start drafting before they know what the story is.
A story that feels alive usually has four things working underneath:
- A pressure system – something forces a decision
- A voice with rules – not an accent, but a way of paying attention
- Characters who hide things – internal logic that creates behaviour
- Earned omission – you know the submerged material, so leaving it out creates weight instead of vagueness
This guide shows how to build those elements before you draft. Each section includes a ready-to-use prompt.
One warning upfront: this method prevents technical bloodlessness: the empty prose that mimics story shape but goes nowhere. It won’t fix emotional bloodlessness. That requires you to write about something you actually care about. No method can save a story you’re not invested in.
Arc: “A Strategy That Stops Working”
Story arc isn’t a plot diagram. It’s a mechanism:
Someone has a way of coping → pressure arrives → the strategy fails → they adapt, change, or break.
Kurt Vonnegut’s “shapes of stories” idea is useful as a sanity check: pick a simple curve so you’re not inventing plot blindly.
John Yorke’s structure work makes the same point: turning points exist because pressure forces decisions.
Before you draft anything, write three lines:
- Start state (8 words max)
- End state (8 words max)
- Pressure (one sentence)
If you can’t articulate those three elements clearly, you don’t have a story yet. You’ve got a situation or a mood.
PROMPT (copy/paste):
Give me 8 possible short-story arcs as plain sentences.
Each arc must include: start state, pressure, irreversible turn, end state.
No lore. No backstory. No theme explanations.
Use short, concrete words.
Format as a numbered list.
Premise: A Sentence That Contains Conflict
Theme is commentary. Premise is machinery.
Lajos Egri’s core framing: premise + character opposition = rising pressure. The premise isn’t what the story “means”; it’s what forces decisions.
Use this template:
When [EVENT] happens, a person who believes [BELIEF] must [ACTION], which costs them [PRICE].
Two guardrails (these stop “AI mush”):
- BELIEF must be something the person could actually say out loud
- PRICE must be something you can show on the page (observable consequence)
PROMPT (copy/paste):
Turn this pressure into 10 one-sentence premises: [PASTE YOUR PRESSURE SENTENCE]
Each premise must:
- imply a hard choice
- show a stated belief colliding with an external demand
- give a price that can happen on the page (observable consequence)
Plain language only. No abstract theme talk.
Voice: A Rule-Set, Not a Costume
Beginners try to get voice by copying a favourite writer. That produces pastiche.
George Orwell’s rules are a blunt antidote: short words, cut dead words, avoid prefabricated phrases, prefer clarity over “writer voice.”
So don’t ask AI to “write like Hemingway.” Instead, define a Voice Charter: five enforceable rules that control attention and prose behaviour.
One guardrail: if you can’t enforce a rule in every paragraph, cut it.
PROMPT (copy/paste):
Ask me 8 yes/no questions about my preferred voice:
- Sentence length (short / varied / long)
- Humour (present / absent)
- Tenderness (direct / withheld)
- Violence (on-page / implied / absent)
- Explicitness (body / emotion / neither)
- Distance (close / remote)
- Pace (compressed / lingering)
- Metaphor (frequent / rare)
Then create a 5-rule Voice Charter based on my answers.
Ensure the rules are mutually reinforcing and enforceable in every paragraph.
Then give two lists:
- 10 DO examples (sentences that follow the rules)
- 10 DON’T examples (sentences that break them)
General constraints: use short words where possible, cut unnecessary words, avoid prefabricated phrases, avoid “writerly” flourishes.
Character: Build the Mask, Not the Biography
This is where AI stories become indistinguishable mush: every character sounds like the narrator, because no one has internal logic that differs from the prose voice.
E.M. Forster’s test: a “round” character can surprise in a convincing way. That “convincing” part is key — surprise must come from private logic.
So don’t write biographies. Build masks:
- What they’re hiding (this can be small: envy, cowardice, compromise)
- How they hide it (strategy)
- What they won’t do (and make that clash with the pressure)
Bad differentiation: three characters who all “hide behind professionalism.”
Good differentiation: one who over-explains rules, one who deflects with jokes, one who goes silent and stares.
PROMPT (copy/paste):
Create Character Mask Sheets for these roles: [LIST 2–4 ROLES]
For each character, answer:
- What do they want today?
- What will be revealed if they lose control?
- What won’t they do, even under pressure?
- How do they avoid exposure when someone gets close?
- What story do they tell themselves to feel like a good person?
- What do they notice first in a room?
- What object or habit gives them away?
Add 3 actor-playable tells (observable behaviour only).
Make each character’s mask + strategy visibly different from the others.
No backstory dumping.
Omission Must Be Earned
Hemingway’s iceberg principle gets misused constantly. People think it means “be vague.” It doesn’t.
It means: you know the submerged material, so when you omit it, the reader feels its weight. If you omit because you don’t know, you create hollow places.
The “submerged material” has two layers:
Layer 1 (Worldbuilding infrastructure): how the system actually works, what enforcement does, why things happen off-stage. This creates felt reality — the reader senses there’s more beneath the surface.
Layer 2 (Character knowledge & stakes): what each character privately fears, what would happen if secrets were spoken, why no one speaks them. This creates tension.
Build both layers as private notes before you draft. Then omit them on purpose.
PROMPT (copy/paste):
Write the hidden layer of this story as private notes (not for publication):
WORLDBUILDING LAYER:
- what the system actually does (not its official purpose)
- how enforcement really works (informal mechanisms)
- what happens off-stage that characters reference indirectly
CHARACTER LAYER:
- what each character is privately afraid of
- what would happen if someone spoke the secret aloud
- why no one does
Keep it concrete and procedural. No metaphors. No thematic summary.
Then write one drafting rule:
“The story will not state the hidden layer directly.”
World-Building: Artefacts, Not Description
If your world exists only in descriptive passages, it feels like a dream. Artefacts make it real.
Flannery O’Connor’s point: fiction convinces through the senses. The reader has to experience it, not be told about it.
In AI terms: stop worldbuilding with explanation. Build the world out of textual artefacts. documents force clarity and can’t hide behind pretty fog.
PROMPT (copy/paste):
Create 8 artefacts from this world. Output ONLY the artefact text: no commentary or explanation.
Include:
- one official notice
- one sticker with microtext
- one community guideline leaflet (bullets)
- one schedule or time-slot grid
- two SMS reminders
- one checklist or intake form
- one scripted “customer service” snippet (6 lines of “kind control” language)
Each artefact must contain:
- one polite euphemism (for something harsh)
- one quiet consequence (buried in logistics)
Drafting + Revision: Allow a Rough First, Then Cut Hard
This is where established advice stops being inspirational and becomes operational:
Anne Lamott: allow a rough first draft so you’ve got something to shape.
Stephen King: adverbs and intensifiers are often fear in disguise; replace them with stronger verbs or concrete detail.
Elmore Leonard: if it sounds like writing, rewrite it; leave out the parts readers skip. Original NYT source (archive reference).
One final element before you draft: the control word. This is a term authorities in your world use to make coercion sound voluntary: “encouraged,” “invited,” “recommended,” “voluntary compliance.” It should appear sparingly (2–4 times), always in the mouth of power, always as polite threat.
PROMPT (copy/paste):
Write the short story using:
- the Arc (start/end/pressure)
- the Premise
- the Voice Charter
- the Character Mask Sheets
- the Hidden Layer rule (never state it directly)
- weave 3–5 artefacts into the story as encountered text
Constraints:
- plain language; short words; cut dead words
- show emotion through action, objects, and silence. Don’t explain feelings
- no workshop vocabulary (“protagonist,” “stakes,” “theme”)
- use a control word 2–4 times (a polite term authorities use for coercion — “encouraged,” “invited,” “recommended” — always in authority/social-control contexts)
- target length: [SPECIFY RANGE]
Then revise in 3 silent passes:
- Pass 1: cut ~25% without losing clarity or emotional impact
- Pass 2: remove adverbs/intensifiers; replace with stronger verbs or concrete detail
- Pass 3: rewrite anything that sounds “writerly”; remove parts readers skip
Output: Title + story text only. No preamble.
The Pre-Draft Gate (Don’t Start Yet)
You’re ready to draft only when you have:
- Arc (3 lines)
- Premise (1 line)
- Voice charter (5 enforceable rules)
- Mask sheets (each key character)
- Hidden layer notes
- Artefact pack
If you’re missing one, don’t draft yet. Build the missing piece.
For shorter work (under 1,500 words): prioritise arc + voice + one strong mask. For longer stories, build all six.
Why This Order Matters
Bloodless fiction isn’t a prose problem. It’s a preparation problem.
Build the engine first. Then let the AI help you assemble the machine.
If the story still feels bloodless after this prep, the missing ingredient is usually a private stake: something the author genuinely fears, envies, misses, or can’t forgive.
