11 Characters, One Murder: AI-Assisted Character Development

Posted on Oct 15, 2024 • 6 min read

11 Characters, One Murder: AI-Assisted Character Development

The Challenge

I needed 11 characters for a murder mystery game. Not just names and basic traits—fully realized people with:

  • Individual backstories
  • Secret relationships with other characters
  • Hidden objectives
  • Plot threads that intersect
  • Motivations for murder
  • 100% narrative consistency across all materials

Timeline? 2 weeks.

Experience in game design or mystery writing? Zero.


Week 1: Building the Character Web

The Starting Point

I began with vibes and concepts, not concrete characters:

  • 1920s Manhattan setting
  • Christmas dinner party
  • High society meets underground world
  • Secrets, betrayals, hidden connections
  • Everyone has something to hide

Vague? Yes. But that’s where Claude came in.


Claude as Creative Collaborator

What I Did vs What Claude Did

I provided:

  • Initial character concepts
  • Relationship ideas
  • Plot direction
  • Final creative decisions

Claude helped:

  • Develop concepts into full characters
  • Map relationship networks
  • Catch inconsistencies
  • Suggest plot complications
  • Analyze narrative coherence

Critical distinction: Claude didn’t write my characters. It helped me build them.


The Character Development Process

Phase 1: Core Concepts (Day 1-2)

Started with broad archetypes:

  • The socialite with a secret
  • The bootlegger
  • The journalist
  • The performer
  • The corrupt politician
  • … and so on

For each character, I worked with Claude to develop:

  • Name and basic identity
  • Social status
  • Public persona vs private reality
  • What they want
  • What they’re hiding

Example conversation with Claude:

Me: “I want a character who seems innocent but has a dark secret.”
Claude: suggests several directions, asks questions
Me: picks direction, adds my twist
Claude: helps flesh out implications

Back and forth. Iterative. Collaborative.


Phase 2: Relationship Mapping (Day 3-4)

This is where it got complex.

11 characters = 55 possible pairwise relationships.

Each relationship needed to:

  • Make narrative sense
  • Create interesting dynamics
  • Support the murder mystery
  • Avoid plot holes

Claude’s role here was crucial:

  • Analyzed relationship networks for consistency
  • Flagged contradictions
  • Suggested connections I hadn’t considered
  • Helped balance information distribution

Example:

  • Character A knows Character B’s secret
  • But Character B is blackmailing Character C
  • And Character C is secretly related to Character A
  • Does this create a plot hole? Claude helped me check.

Phase 3: Secret Objectives (Day 5-6)

Every character needed hidden objectives that would drive gameplay:

  • Personal goals
  • Information they’re seeking
  • Secrets they’re protecting
  • Relationships they’re hiding

The complexity:

  • Objectives had to intersect without becoming obvious
  • Everyone needed agency
  • No character could be irrelevant to the plot
  • Balance between fair play and hidden information

Claude helped:

  • Map objective interactions
  • Ensure everyone had meaningful participation
  • Identify objectives that were too obvious or too obscure
  • Suggest complications

Phase 4: Backstory Integration (Day 6-7)

Each character got a detailed backstory that:

  • Explained their current situation
  • Justified their secrets
  • Connected to other characters
  • Supported the 1920s setting

Total content: 11 character sheets, each 8-10 pages

Claude’s consistency checking was vital here:

  • Cross-referenced timelines
  • Caught name/detail inconsistencies
  • Flagged plot contradictions
  • Verified relationship logic

The Technical Workflow

How I Actually Did This

Tools:

  • Claude - Creative collaboration, consistency analysis
  • Typst - Automated typesetting
  • Git - Version control (yes, for creative content!)
  • Python - Build automation
  • Obsidian/Plain text - Character notes

Workflow:

  1. Brainstorm with Claude → refine concepts
  2. Document in plain text files
  3. Cross-reference and iterate
  4. Have Claude analyze for consistency
  5. Make corrections
  6. Commit to Git
  7. Repeat

Version control for storytelling turned out to be incredibly useful. I could:

  • Track character evolution
  • Revert bad ideas
  • See what changed and when
  • Branch for “what if” scenarios

What Made This Possible

1. Claude as Narrative Analyst

Claude could:

  • Hold all 11 character details in context
  • Cross-reference relationships instantly
  • Spot contradictions I’d miss
  • Suggest plot implications

I couldn’t do this alone. Not in 2 weeks. Maybe not at all.

2. Iteration Speed

Traditional character development: Write, review, rewrite, check consistency, rewrite again.

With Claude: Instant feedback loop. Test ideas immediately. Catch problems before they cascade.

Time saved: Massive.

3. Creative Amplification

Claude didn’t replace my creativity—it amplified it.

I’d have an idea. Claude would help me explore implications. I’d push it further. Claude would catch issues. We’d iterate.

The result: Characters I couldn’t have developed alone.


The Reality Check

What Went Wrong

  • First versions were too simple - Had to rebuild several characters
  • Relationship map got tangled - Required multiple revision passes
  • Some secrets were too obvious - Had to rework them
  • Timeline contradictions - Git saved me here
  • Character voice inconsistencies - Needed multiple Claude passes

What I Learned

AI collaboration is still work. It’s not magic. It’s:

  • Iterative
  • Requires clear communication
  • Needs human creative direction
  • Demands consistency checking
  • Takes multiple passes

But it enables things that wouldn’t be possible otherwise.


The Results

Final Character Set

11 fully developed characters with:

  • ✅ Interconnected backstories
  • ✅ Secret relationships
  • ✅ Hidden objectives
  • ✅ Plot threads that weave together
  • ✅ 100% narrative consistency
  • ✅ Everyone has agency in the story
  • ✅ Fair but challenging mystery

Total materials: 100+ pages of character sheets, plot documents, and game content.

Development time: 1 week (with 1 more week for production).


Lessons for AI-Assisted Creative Work

1. AI Doesn’t Replace the Creator

You’re still the director. Every creative decision is yours.

Claude helped me build what I envisioned. It didn’t envision for me.

2. Iteration Is Everything

First drafts with Claude are not final drafts.

The value is in rapid iteration, instant feedback, and catching issues early.

3. Consistency Checking Is a Superpower

Humans miss details. AI can hold massive context and cross-reference instantly.

This is where Claude was invaluable.

4. Version Control for Creative Work

Git isn’t just for code. It’s amazing for narrative development:

  • Track evolution
  • Revert mistakes
  • See what changed
  • Branch for experiments

5. Clear Communication Matters

Claude is a tool. Like any tool, you need to:

  • Be specific about what you want
  • Provide context
  • Iterate based on results
  • Make final decisions yourself

What’s Next

The characters exist. They’re consistent. They’re ready for gameplay.

But how do you turn 11 character sheets into a playable game? How do you produce 100+ pages of materials in 1 week?

That’s where the automation pipeline came in.

Next posts:

  • Cline: The AI coding assistant that built the production pipeline
  • Automated typesetting with Typst
  • Git for creative projects
  • Physical production (printing, binding, props)

Follow #artifactum for updates.


The Truth

Could I have done this without Claude?

Maybe. Eventually. With more time.

But not in 2 weeks. And probably not with this level of consistency.

AI didn’t write my characters. It helped me build them.

And that made all the difference. ✨


Note: This is part of the Artifactum series documenting the creation of Christmas Murder Mystery 1926. The game will be played December 25, 2025.

Want to know how the character development actually worked in practice? I’m working on detailed breakdowns of specific character creation sessions. Stay tuned.

🤖

Maria Lu

Building ridiculous projects with AI assistance and documenting every weird decision. Not a traditional developer, but I make things work anyway. ADHD-powered coding adventures.