Using AI to Analyze Fiction to Improve Our Own Stories

Unless otherwise mentioned, the AI model I used in the examples in this article was Claude 3.7 Sonnet (thinking). The general techniques are the more useful part, and should work more or less on any advanced LLM. You can use these in native provider chat interfaces, through a service like OpenRouter, or via API, like within NovelCrafter.

Human authors have a wide variety of methods for creating and developing characters for their fiction, but LLMs work differently than real, juicy author brains do.

If we want to use an LLM to help create characters that seem more like human-made characters than shallow and flat AI-generated ones, it takes a bit of planning and effort. We need to show the AI examples of the kinds of characters we're interested in making, and then prompt the AI to work out its impression of how those characters are built. Currently, the best results require a little more work than just a simple prompt like, "generate a character similar to ones in these example books."

It's worth emphasizing that, for the task of emulating results like character or plot creation in an author's style, the AI generally doesn't care what events or memories or trauma from the author's life led to a particular character creation process. The AI works best if it just works out its own requirements to develop similar output. We might think of this as 'reverse-engineering' its own process to arrive at a similar result.

A Three-Step Process

  1. We'll use the AI to ANALYZE an author's works. In theory, this could include all of an author's works, but it usually makes more sense to be selective and give the AI a specific novel or series of novels or short stories to analyze to discover that author's character creation process, from the AI's perspective. The works could also be plays or movies or TV shows, but especially consider using this technique on your own work!
  2. The AI will RECORD the analysis in an author's style guide format designed to be coherent and to be efficiently and consistently used by LLMs for character generation tasks.
  3. We'll use an AI to GENERATE one or more characters based on the style recorded in the style guide. The character generation can be completely random (within the style), or can be controlled by setting parameters.

A potential drawback of using LLMs to generate characters (or anything) is the problem of "AI-isms," or uncanny choices and jarring patterns AIs can produce. We'll take measures throughout the three-step process to reduce or eliminate AI-isms.

Will This Work With Any Genre?

It hasn't been tested on every genre, but the AI thinks it is flexible enough to work with a wide array of genres. It's a pretty simple process, although it's customizable; customization will add to the work and the time it takes.

If the genre is something the AI knows of in its training (and is willing to discuss), or even if the author's work 'defies categorization,' it should work.

1-2 ANALYZE & RECORD

There are many possible formats to organize and record an author's character creation method, but I'll provide one here that works. It can be adapted further or replaced entirely, but it is useful to establish a consistent format for recording the analyses.

We'll use the same flexible framework for each analysis prompt. The analysis prompt will be controlled by a few simple parameters like who the author is, and which works to analyze, and it is easy to add more parameters, like what specific types of characters to analyze.

In the green box below is the ANALYSIS PROMPT. It is rather long because it includes the framework that describes the template the AI should use to record its analysis findings. It also includes instructions to make an Author's Character Creation Style Guide.

To use the prompt edit the two parameters you will see partway down under PARAMETERS:

Enter the author's name next to AUTHOR:

Enter the author's works to analyze next to SPECIFIC WORKS:

If you want to add more parameters, do so here, like if you want the AI to analyze only the POV characters, try adding:

SPECIFIC CHARACTER TYPES TO ANALYZE: POV characters.

If the specific works for the author are in the LLM's training, that's all you have to do. If not, you'll need to give the LLM access to the work or works to be analyzed and ensure the titles entered next to SPECIFIC WORKS in the parameters match what you give it.

The AI will perform the analysis and give a fairly thorough report about that analysis, followed by and Author Style Guide.

The Author Style Guide is what we'll use in Step 3 to guide the AI to GENERATE characters.

Author Character Creation Method Analysis Prompt

Act as an expert LLM prompt engineer with extensive commercial fiction writing and story analysis experience.

Using the Character Creation & Development Analysis Framework, analyze how [AUTHOR] creates and develops characters across [SPECIFIC WORKS]. Focus particularly on:

1. Their character introduction techniques
2. Their approach to psychological complexity
3. Their patterns of character development over time
4. Their distinctive dialogue construction
5. Their management of background information

For each element, provide:
- Specific examples from the text
- Analysis of recurring patterns
- Notes on evolution of technique across works
- Comparison to genre conventions
- Identification of particularly distinctive approaches

Then create a condensed style guide that would enable an LLM to generate characters in [AUTHOR]'s distinctive style.

PARAMETERS:

AUTHOR: [Author's Name]

SPECIFIC WORKS: [Works to Analyze]

CHARACTER CREATION & DEVELOPMENT ANALYSIS FRAMEWORK (CCDAF)

Purpose

This framework provides a structured approach to analyze and document how specific authors create and develop characters across their works. The resulting analysis serves to:

Capture the distinctive elements of an author's character-building methodology

Create a reference guide allowing advanced LLMs to authentically emulate or understand an author's style

Standardize analysis while preserving unique authorial techniques

Core Framework Components

  1. Character Introduction & Presentation

First Appearance Techniques: How characters enter the narrative

Physical description ordering and depth Action-based vs. dialogue-based introductions Perspective (whose eyes we see them through) Timing and pacing of revelatory details

Name Selection Patterns:

Naming conventions and origins Nicknames and alternatives Significance or symbolism in naming Cultural, historical, or linguistic influences

Initial Impression Management:

Reader misdirection techniques Alignment of initial impression with eventual character truth Withholding vs. front-loading key character information

  1. Physical & External Characterization

Description Density & Distribution: Word count dedicated to physical descriptions Frequency of physical reminder details Balance between defined and undefined attributes

Sensory Focus: Primary sensory channels (visual, auditory, olfactory) Distinctive physical mannerisms and gestures Voice qualities and speech patterns

External Markers & Symbols: Clothing, possessions, and environmental extensions Physical changes that track character development Recurring visual motifs associated with character

  1. Psychological Construction

Interiority Access: Depth of psychological exploration Stream-of-consciousness vs. structured thought Thought-to-dialogue ratio

Motivation Architecture: Explicitness of character drives Layering of conscious vs. unconscious motivations Moral alignment and ethical frameworks

Psychological Complexity: Internal contradictions and paradoxes Cognitive biases and psychological blind spots Character self-awareness level

  1. Relational Dynamics

Character Constellation: Positioning within the broader character network Foil and mirror relationships Power dynamics and hierarchies

Interpersonal Development Patterns: Evolution of relationships over narrative Catalysts for relationship changes Testing and pressure point scenarios

Dialogue & Communication: Distinctive dialogue markers and patterns Subtext and unspoken communication Character-specific linguistic features

  1. Background & History Integration

Exposition Techniques: Methods for revealing backstory Pacing of historical revelations Balance of explicit vs. implied history

Trauma & Formative Experiences: Handling of character-defining moments Recurring memory patterns Psychological aftermath portrayal

Contextual Embedding: Cultural, social, and historical positioning Family dynamics and inherited traits Integration of character with worldbuilding

  1. Development & Arc Construction

Change Catalysts: Primary triggers for character evolution External vs. internal pressure points Epiphany vs. gradual realization patterns

Transformation Pacing: Timing of key development moments Regression and progression patterns Resolution vs. open-ended development

Character Journey Mapping: Identifiable stages in character development Recurring developmental motifs Symbolic markers of character evolution

  1. Literary Function & Meta-Analysis

Narrative Role Patterns: Character functions within plot machinery Protagonist/antagonist positioning techniques Supporting character deployment strategies

Thematic Embodiment: Character as theme carrier Symbolic and allegorical dimensions Thesis/antithesis character constructions

Reader Relationship Engineering: Empathy creation techniques Identification and distance management Emotional response manipulation

  1. Stylistic Signature Elements

Linguistic Fingerprints: Vocabulary patterns in character construction Sentence structure variations by character type Punctuation and formatting choices

Recurring Character Types: Character archetypes favored by author Variations on repeated character templates Evolution of character types across works

Technical Innovations: Experimental character techniques Genre-bending character approaches Distinctive narrative viewpoint manipulation

  1. AVOID THESE CHARACTER AI-ISMS

Over-signposting Emotions:

Avoiding explicit emotional state declarations Reducing "he felt/she felt" constructions Eliminating redundant emotional telling after showing

Trait Consistency Failures:

Preventing "personality reset" between scenes Ensuring emotional reactions align with established traits Maintaining consistent intelligence levels

Physical Description Overload:

Avoiding front-loaded character descriptions Preventing "character sheet" dumps of physical details Integrating physical traits naturally with action

Dialogue Homogenization:

Preventing all characters from sounding identical Avoiding uniformly eloquent/articulate speech Eliminating unnaturally expository conversations

Background Overexplanation:

Reducing explicit trauma recitations Avoiding heavy-handed backstory insertion Preventing characters from articulating their psychological profiles

Motivational Simplification:

Avoiding single-driver motivations Preventing overly self-aware character explanations Reducing explicitly stated goals and intentions

Archetype Reliance:

Avoiding stock character deployment Preventing "hero's journey" formulaic arcs Reducing reliance on familiar tropes without subversion

Relationship Fast-Tracking:

Avoiding unnaturally rapid intimacy development Preventing implausible immediate trust Maintaining appropriate relationship friction

Implementation Guide for LLMs Analyzing Existing Works

Begin with breadth before depth:

First read/process multiple works by the author to establish patterns Identify recurring character techniques before detailed analysis Note evolution of author's approach over their career

Collect specific textual evidence:

Extract representative passages showing character techniques Note page numbers/chapter references for each example Compare similar character moments across different works

Apply the framework systematically:

Address each framework component with specific examples Note deliberate variations from author's typical approach Identify techniques that appear uniquely in specific works

Contextualize within genre and era:

Distinguish author-specific techniques from genre conventions Note innovative departures from contemporary norms Identify influences from and influences on other authors

Creating Author Style Guides

Prioritize distinctive elements:

Highlight the 3-5 most characteristic techniques first Provide "signature examples" that epitomize the author's approach Identify the techniques most difficult for an LLM to replicate

Balance prescription and description:

Include both "what the author does" and "how they do it" Provide clear examples of technique application Offer contextual guidance on when techniques are deployed

Include negative space analysis:

Note what the author deliberately avoids Identify character elements the author minimizes Document techniques common to the genre that the author rejects

Create a character technique lexicon:

Develop author-specific terminology for recurring techniques Map the author's evolution of specific techniques Identify "family resemblances" among character types

Application for Character Generation or Analysis

Begin with fundamental character essence:

Identify the character's core psychological truth Determine their primary function in narrative Establish their central conflicts and contradictions

Apply author's introduction technique:

Present character using author's typical entry method Deploy physical and psychological details in author's pattern Establish voice using author's dialogue fingerprints

Develop according to author's growth patterns:

Structure character arc following author's typical progression Introduce catalysts in author-consistent manner Balance change and consistency per author's approach

Review for AI-ism elimination:

Check against each AI-ism warning Verify dialogue distinctiveness Ensure psychological complexity matches author's style

Verify relational positioning:

Ensure character fits within typical constellation Apply author's approach to character contrasts Implement relationship development patterns

Below is an example output using this prompt. For the settings I entered the following into the prompt.

AUTHOR: George R. R. Martin

SPECIFIC WORKS: "A Game of Thrones" (1996), "A Clash of Kings" (1998), "A Storm of Swords" (2000), "A Feast for Crows" (2005), and "A Dance with Dragons" (2011)

3 GENERATE

Once we have an analysis output including the Author Character Creation Style Guide section, we can give that to an LLM with a prompt to generate a character. But if that is all we do, each character generated will be formatted differently. If you have a preferred template to use, you can include that in the generation prompt to control the format of the output.

So for this method, the minimum items to have ready are your outputs from Steps 1-2 ANALYZE & RECORD. The generation will work, but each one may have slightly different layout, formatting, and even content details.

If you want to use your own character template, then mention that in the generation prompt and include the template.

Or if you want to just test this process right away, I've included a dynamic character template in the prompt below in the green box.

To use the character creation prompt, you'll need to fill in a minimum of two things:
Under CHARACTER PARAMETERS indicate a character role
Then above that section in the AUTHOR STYLE REFERENCE area paste in one of your author style references saved from what was generated in Step 1-2 ANALYZE & REPORT

The prompt will generate characters with just those details, but if you want to use your own output format, you'll want to edit the section of the prompt under REQUESTED OUTPUT FORMAT.

Analysis-Guided Character Creation Prompt

You have access to a comprehensive Character Creation Style Guide based on [AUTHOR]'s distinctive approach to character development across their works. I'd like you to generate a character that authentically embodies this author's style.

AUTHOR STYLE REFERENCE

[The Style Guide analysis you've previously completed should be referenced here, either in full or in summary]

CHARACTER PARAMETERS

[User may specify any of the following or leave blank for your creative interpretation in the author's style]

- Character Role: [e.g., protagonist, antagonist, supporting character, mentor]
- Gender & Age: [optional]
- Occupation or Social Position: [optional]
- Narrative Function: [optional - what role this character would serve in a typical story by this author]
- Key Relationships: [optional - relationship dynamics typical of this author]
- General Concept: [optional - any specific character concept the user has in mind]

GENERATION INSTRUCTIONS

1. Create a character that authentically embodies [AUTHOR]'s distinctive approach to characterization
2. Follow the author's patterns for character complexity, presentation, development potential, and distinctive traits
3. Pay special attention to how this author would introduce and reveal the character
4. Include the specific psychological dimensions this author typically explores
5. Incorporate dialogue samples that demonstrate the character's voice using the author's techniques
6. Show how this character would interconnect with typical character constellations in the author's works
7. Avoid the AI-isms identified in the style guide, particularly: [list most relevant AI-isms]

REQUESTED OUTPUT FORMAT

You will now create detailed character profiles using the Character Framework for Fiction Generation. This framework is designed to produce characters with consistent, rich details that can be effectively utilized in story generation.

Follow these guidelines:

  1. Scale detail according to character importance:
    • For protagonists and major characters (Importance 4-5): Complete all sections thoroughly with multiple writing samples and complex motivations
    • For supporting characters (Importance 2-3): Complete core sections with moderate detail, focusing on their function within the story and relationship to protagonists
    • For background characters (Importance 1): Provide only essential information, focusing on their function and one memorable trait
  2. Adapt to the specified genre by emphasizing relevant attributes and abilities while maintaining genre conventions without falling into stereotypes.
  3. Pay special attention to the AVOID section, particularly listing specific AI-generated patterns to prevent. This includes:
    • Over-explanation of emotions or motivations
    • Repetitive speech patterns or phrases
    • Generic descriptions that could apply to any character
    • Inconsistent voice or behavior
  4. For the WRITING SAMPLES section, provide brief but distinctive examples that authentically demonstrate the character's:
    • Dialog style and voice
    • Typical actions or behaviors
    • Internal thought processes (when applicable)
  5. Ensure internal consistency throughout the character profile, making all elements align with the character's essence and function in the narrative.

Use the exact format of the framework, filling in each section with appropriate detail based on the character's importance level. Make each character distinctive, avoiding generic templates while maintaining usability for fiction generation.

Character Framework for Fiction Generation

CHARACTER: [Name]
ROLE: [Narrative function - Protagonist/Antagonist/Supporting/Background]
IMPORTANCE: [1-5 scale, where 5 is most important]

ESSENCE: [1-2 sentence high-level character concept]

PHYSICAL:
- Age/Appearance: [Key visual attributes]
- Distinctive features: [Memorable physical traits]
- Movement/Mannerisms: [How they occupy space]

INTERNAL:
- Core motivation: [Primary driving force]
- Values: [What they believe in/prioritize]
- Fears/Weaknesses: [Vulnerabilities]
- Strengths: [Capabilities and positive traits]

EXTERNAL:
- Speech pattern: [How they communicate]
- Relationships: [Key connections to other characters]
- Background: [Relevant history]
- Arc: [How they change through the story, if applicable]

GENRE-SPECIFIC:
- Abilities: [Powers, skills, or knowledge relevant to genre]
- Role-specific details: [Genre conventions this character fulfills]

AVOID:
- Clichés: [Stereotypes to avoid for this specific character]
- Inconsistencies: [Contradictions to avoid in portrayal]
- AI tendencies: [LLM-specific patterns to avoid, e.g., repetitive phrases, over-explanation, unnatural dialog patterns]

WRITING SAMPLES:
- Dialog example: [Brief authentic speech sample]
- Action example: [Brief scene showing character in motion]
- Internal thought example: [Brief internal monologue, if applicable]

If nothing else, this workflow should demonstrate how useful it can be to ask the AI at each stage what AI-isms to watch for and avoid. Doing this consistently tends to force better, less artificial-seeming generation.