Building Believable Characters

Every character you’ve ever loved in fiction was built around a loss.

Not a vague sadness. Not a rough childhood you vaguely reference. A specific thing that happened, or a specific thing that was taken, or a specific thing they’re terrified of losing next. The rest of the character radiates from that.

Harry Dresden - lost his mother at birth, spent his teen years being used by the man who was supposed to protect him. Every case he takes is a man trying to prove he won’t abandon someone the way the world abandoned him.
Ender Wiggins - isolated by design, weaponized by adults, then handed a genocide he didn’t know he was committing. The rest of his life is the weight of that.
Kinsey Millhone - orphaned at five, raised by an aunt who kept her at arm’s length. She builds the same walls in every relationship and wonders why she’s alone.
Jack Reacher - gave everything to an institution that handed him nothing on the way out. No home. No possessions. No roots. A man who decided that if nothing is his, nothing can be taken.

Every one of them was designed around what they lost, not what they do.

That’s the whole job. Design the wound, then design the person who built a life around it.

A Note for Pantsers

Writing circles are divided into two camps. Plotters plan before they write. Pantsers discover the story as they go, by the seat of their pants, with no map and no apologies.

This framework is not entirely compatible with pantsing. I’m sorry. You are going to need to do some planning to make this work.

Not a forty-page outline. Not a complete story bible before chapter one. But AI needs to know who your characters are before it can help you write them, and that means you need to know too. A small amount of upfront thinking now prevents a large amount of confused output later.

Pantsers, you can still walk up the staircase with a candle. But you do need to know your characters, even if the story is evolving as you go.

Start With a Human

Before you write a single scene, before the plot exists, before the magic system is finished - you need to know who your people are. Not what they do in the story. Who they are.

A character is not a role. A character is a person who happens to fill a role, and those two things feel completely different on the page.

Here’s the framework I use for every new character:

Start with a human. Whatever species, whatever world - they have to be recognizably human in their psychology. Readers don’t connect with vampires. They connect with loneliness wearing a set of fangs.
No one is the villain of their own story. Your antagonist thinks they’re right. The moment you let a character become a symbol of wrongness instead of a person with a perspective, the reader checks out.
Name, species, date of birth. The basics. They anchor everything else.
Backstory. Not their whole life history. The part that explains why they flinch at certain things, why they’ve built the walls they’ve built.
Ambitions. What do they want? The external goal, yes. But also the deeper thing underneath it.
Convictions. What do they believe is true about the world? This is where characters start to argue with each other in ways that feel earned.
Enneagram or Archetype. A shorthand label that keeps you consistent. Not a cage. A compass.

That last note matters: these are compasses. Characters grow. The framework is a starting point, not a sentence.

Something to Lose

Every character needs something to lose, or something already lost.

A character with nothing at stake is a function in a plot. A character with something to lose is a person.

The moment your reader understands what a character is carrying - the grief, the ambition, the thing they’re protecting - they’re invested. Pull that thread away and you have a story that happens to people, not a story driven by them.

I call the formative event the Key Wound. It’s the thing that explains why this person moves through the world the way they do. Your antagonist’s Key Wound is what makes them not the villain in their own head. Your protagonist’s Key Wound is what makes their arc feel earned when they finally work through it.

Take Betsy Bishop, a character from my own series, Vampires of Tucson. She was born in Salem in 1676. Her mother was accused of witchcraft and burned at the stake while Betsy was still a teenager - close enough to see the smoke, far enough away that being found meant dying beside her.

She ran. A couple of centuries and a vampire embrace later, she lost Sings-In-Woods, her oldest friend and the one who turned her. Sings took her own life in 1874, a few weeks after the man she’d spent years fighting beside died of natural causes.

Those two losses are the load-bearing walls of Betsy’s entire character. The first one taught her that the world is not fair and sometimes survival requires distance. The second broke through every wall she’d built in the two hundred years between. Every decision she makes in the present-day story runs through one of those two events.

That’s the Key Wound in practice. You don’t need a page of backstory. You need the one or two events that explain why this person became this person.

Don’t skip it. Don’t leave it vague. Write it down.

AI Needs Structure

Here’s where it gets practical.

When I started this phase of the project, I had no AI experience and high expectations. What I also had was a collection of beautifully formatted character PDFs - portrait, consistent layout, all the information any reader would need.

ChatGPT choked on them. Claude choked on them. My Word documents fared no better.

The format I’d spent months building was completely useless to AI. So I first had to adjust my expectations.

At that point I had two options: I could try to force AI to work with what I had, or wait until it was capable of using them. Or I could bring the mountain to Mohammad.

Computers do not read the way humans read. Humans want visual hierarchy, portraits, nice headers, white space. Computers want rows and columns.

Give a person a comma-delimited file and they will hate you for it. Not for a day. For life. The reverse is not true for AI - structured data is exactly what it wants.

So the answer, annoying as it is to hear, is a spreadsheet. Build your characters in Excel or Google Sheets. Not as your primary creative workspace - you can still write your beautiful character bibles in whatever format you love. But your AI-facing data needs to live in a structured format: consistent column names, one character per row, no merged cells, no portraits, no decorative formatting.

The AI cannot appreciate the portrait. It can, however, tell you exactly what your character would do in chapter seventeen if you’ve given it the right columns.

The Schema

Here’s the header row. Drop it directly into Excel or Google Sheets, or save it as a pipe-delimited file and import it.

One note before you look at it: my own schema is significantly different from this one. It has been specialized for vampire fiction over years of iteration - fields like mortaldays, vampirenights, aidialogueprompt, and aiimageprompt that don’t translate to a general audience. What follows is the most universally useful version, designed to work for any story, any genre, any species. Start here, then adapt it to your world.

Character_ID|Full_Name|Known_As|Species|DOB|Age_At_Story_Start|Gender|Backstory_Summary|Key_Wound|Ambitions|Core_Convictions|Self_Perception|Enneagram|Archetype|Character_Voice|Role|Status|Notes

A few notes on what matters here:

Character_ID is the anchor. Use something simple: CHAR001, CHAR002. AI uses this to track who you’re talking about across queries. Don’t skip it.

Key_Wound is the formative loss. The event that explains everything else. If you’ve done the framework work above, this column writes itself.

Self_Perception is your “no villain of their own story” column. How does this character see themselves? What story do they tell about their own life? This is different from how other characters see them, and different from how the reader sees them. Keep all three distinct.

Age_At_Story_Start doesn’t need to be typed manually. If your DOB column is formatted as a date, let the spreadsheet do the math. In Excel or Google Sheets: =DATEDIF(E2, DATE(2024,1,1), "Y") - replace E2 with your DOB cell and 2024 with your story’s year. It calculates the exact age automatically.

Known_As handles nicknames, titles, aliases. Keep Full_Name clean.

Character_Voice is the compiled voice architecture: generation, region of origin, first language, class, education, religious upbringing, and any other layers that explain why this person sounds the way they do. Not the dialect markers. The foundation those markers grew from. Think of it as the answer to the question your AI will ask every time it writes dialogue: why does this character use these words, and where did those words come from?

One thing writers resist about spreadsheets: they think it means reducing everything to a few words per cell. It doesn’t.

I’ve been known to joke that any monkey can write a novel. It’s just a matter of teaching it to type “banana” and then having it do so 50,000 times. NaNoWriMo logic.

Each Excel cell holds 5,461 bananas. Google Sheets holds 8,333. The average novel chapter runs between 1,500 and 5,000 words. You can fit an entire chapter in a single cell, with room to spare.

If you have a detailed backstory in Scrivener, a character bible in Word, or years of notes in InDesign, you can copy and paste directly into the relevant column. The structure is the point, not the brevity.

Once your characters are in the spreadsheet, you can pass that row to your AI and it will actually know who you’re talking about. (We will talk PowerShell exports, JSONL, MCP, Databases, and Claude CoWork in a few weeks.) Not approximately. Specifically.

The difference between a well-structured character row and a PDF portrait is the difference between a trained service dog and a dog you’ve described to someone over the phone.

One is working. The other is hoping.

Think of your AI service dog looking up at you with those big puppy eyes, asking one thing: what would your character do, and why?

The spreadsheet is how you teach it the answer.

The Dartboard Test

Once you have a character row in your spreadsheet, here is how you know it’s working.

The “This Is My Human” and Intent documents anchor AI to your creative vision before it ever reads the character data. The schema row gives it the structured facts. The prompt tells it what to do with both. All three together is what produces a dartboard hit instead of a guess.

Your writing prompt for the mirror scene looks like this:

[Attach or paste your “This Is My Human” document here.]

[Attach or paste your Intent document here.]

Character_ID    Full_Name   Known_As    Species DOB Age_At_Story_Start  Gender  Backstory_Summary   Key_Wound   Ambitions   Core_Convictions    Self_Perception Enneagram   Archetype   Character_Voice Role    Status  Notes
CHAR001 Joe Schmo   Joe Human   6/24/1987   37  Male    etc etc etc etc etc etc etc etc etc etc

Write a scene of 300 to 500 words in which [‘name’] stands alone in front of a mirror. They are not admiring themselves. They are confronting something.

The scene must:

1. Open with [‘name’] noticing something in their reflection that connects to their secret pain.

2. Surface the [‘Self_Perception’] about who they are and why things happened the way they did.

3. Reveal what they want [‘ambition’] and the honest reason they are not sure they deserve it.

4. End on their [‘convictions’] - the thing they tell themselves to keep moving.

Voice: First person, present tense. No one is watching. The character is as honest with themselves as they allow themselves to be. Their [‘archetype’] should be recognizable in how they frame the confrontation, not just what they say.

Here is what that prompt produces when I run it against one of my own characters. Betsy Bishop, Vampire, embraced 1697. Her Key Wound is the loss of Sings-In-Woods, her oldest friend and the one who turned her, who Betsy believed took her own life in 1874.

I don’t linger in front of mirrors.

There’s no practical reason to. My face hasn’t changed since 1697, and staring at something you can’t change is a particular kind of punishment. The mirror in my office is there because it came with the building. I keep it because sometimes you need to see the room behind you without turning around.

Tonight I’m looking at my own eyes and thinking about Sings.

It had been a year since she died. A year since the word reached Tucson, and I sat in this same chair and understood for the first time what it means to be alone in a way that three hundred years of living couldn’t fix.

She made me. She stayed. And then she didn’t.

I keep Tucson free. No eastern influences, no outside organization with their fingers in my territory. I’ve held this city through annexation and two world wars and I’ll hold it through whatever comes next.

Most days that’s enough to get me out of this chair.

I’m not sure I earned it. The territory, the title, the right to call this city mine.

She built me into someone capable of holding something this large. She was supposed to be here while I held it. I keep waiting to feel like I deserve what she gave me.

It hasn’t happened yet.

The Pirate Queen of Tucson. That’s what they call me.

I look at the woman in the mirror and I say the only thing that still works: protect what is mine. It got me through Nassau and New Orleans and a hundred years of open prairie and it will get me through this.

It has to. I don’t have another answer.

That is a dartboard hit. The AI knew her wound, her history, and her archetype - because the data was there to know. The prose is clean and competent. It sounds like a vampire novel.

What it doesn’t sound like yet is Betsy. That’s the voice field. We’ll add it next time and run the mirror again.

Betsy’s full character profile is 16,736 words. That’s what’s behind the scene above. You don’t build that in a day, and you don’t need to before you start writing.

But every word you add to those cells is context your AI can use. The more it knows, the less it invents. Or hallucinates.

The voice field and the image field are their own conversations. We will get to how to build both in future posts.

When we get to voice, we will run the mirror prompt again with that field loaded. Then again after we add the image prompt. Three runs, same character, same wound - and you will be able to see exactly what each field contributes.

Working With Your Service Dog

If your AI produces something generic instead of something you can accept, you haven’t given it enough. Go back and iterate until AI can get it on the board.

You can tell AI no. In fact, it is imperative that you do. If the output isn’t meeting what you can accept, say so. Don’t move on.

Don’t settle for something that doesn’t sound like your character just because it’s coherent prose. Tell it exactly what you wanted and why this missed. Ask it what your character bible might be lacking and how to strengthen it.

The service dog does not just fetch - it can help you figure out why it couldn’t find what you sent it for. Use it that way.

I have lost count of the number of times I have asked: “How can we do this better?” “How can I improve the schema?” “How would you phrase this?” “I am looking for this result - how do we get there?”

Every one of those questions made the next output better. The conversation is part of the work.

Make the AI service dog your partner.

Start Here

Start with a human. Give them something to lose. Put it in a spreadsheet. Then teach your service dog what it needs to know.

The framework, the wound, the schema, the conversation - none of it is complicated. It just requires you to do the work before you ask AI to do the work. Writers who skip that step tend to be the ones still getting generic output six months from now, wondering why the tool isn’t performing.

The tool is performing. It just doesn’t know your character yet.

Introduce them.