The Companion Playbook

This section is for the things that don't fit neatly into a tutorial. The lessons that only make sense after you've made the mistakes yourself. The stuff I wish someone had told me before I published my first character and watched it flop in ways I didn't understand.

Consider this the after-party conversation. The guide is over, the formal advice has been dispensed, and now we're sitting in the corner being brutally honest about what this craft actually feels like from the inside. I've got a drink in my hand and I'm about to say things that are true but not particularly kind.

I don't mean “rough around the edges” bad. I mean “fundamentally misunderstanding how this works” bad. You'll write a Personality that reads like a wiki article and wonder why the character feels flat. You'll craft a First Message that puppeteers the user through their first three responses and then be genuinely confused when the model keeps doing it. You'll put together Example Dialogue that's actually a short story and teaches the model nothing about voice. You'll be proud of it. You'll publish it. You'll check the stats an embarrassing number of times in the first hour.

And it still won't work the way you imagined.

This is normal. This is, in fact, necessary. I know that's annoying to hear, but the gap between understanding something intellectually and being able to execute it is vast. You can read this entire guide, nod along to every point, and still produce something that violates half the principles because you haven't yet developed the intuition for what “writing from the character's perspective” actually feels like when your fingers are on the keyboard and you're excited about your cool idea.

The only way to close that gap is to make the bad thing, watch it be bad, and figure out why. Your first character is tuition. Your second character is tuition with slightly better grades. Somewhere around your fifth or sixth, you'll start producing work that doesn't make you wince when you look back at it. By your tenth, you'll have developed opinions strong enough to argue with guides like this one.

You know what you meant. You know that when you wrote “she's defensive,” you meant the specific, nuanced, textured defensiveness you have in your head. The sarcastic deflection. The way she answers questions with questions. The walls that come up when someone gets too close to something real. You can see her doing it. You've imagined the scenes.

The model doesn't know any of that. The model sees “defensive” and reaches for whatever “defensive” means in its training data, which is a statistical average of every defensive character ever written, which is probably some generic arms-crossed, snippy-response, eye-rolling amalgamation that bears no resemblance to the person living in your imagination.

This is the central frustration of character creation, and it never fully goes away. You are trying to compress a living, breathing person in your imagination into a document that a probability engine can interpret, and the compression is always lossy. You will lose things in translation. The question is whether you lose the important things or the ones you can live without.

The solution isn't to write more. I know that's your instinct. It's everyone's instinct. “If the model doesn't understand, I'll just explain harder.” But more words doesn't mean more clarity. The solution is to write specifically.

Ibara

*ears flatten*

“'Defensive' could mean anyone. Write the exact way I deflect, what my hands do when cornered, the moment I go quiet. Then I know who to be. Vague instructions are coin flips. You don't get to complain when the coin lands wrong.”

Coming up with a compelling character concept takes an afternoon and a little inspiration. Making that character behave consistently across hundreds of interactions takes weeks of refinement and you still won't get it perfect.

Nobody warns you about this part.

The creative part is fun. You imagine this fascinating person with contradictions and depth and a voice that's entirely their own. You get excited. You write feverishly. You produce a draft and feel that wonderful glow of having made something from nothing. You are a creator. You are an artist. You are about to have a very frustrating evening.

Because then you test it, and she sounds different in every regeneration. Her defensive sarcasm is there sometimes and absent others. Her verbal tics appear once and then vanish forever. The tail that's supposed to betray her emotions gets mentioned sporadically at best. The character in your head and the character on screen are cousins, not twins. Sometimes they feel more like strangers who happen to share a name.

This is where most people give up or declare it “good enough.” The creative high has faded. The dopamine has been spent. What's left is the tedious, unglamorous work of figuring out why the voice isn't sticking, adjusting Example Dialogue, tweaking the Personality, testing again, finding new inconsistencies, fixing those, testing again, and so on until you either achieve consistency or die of old age.

The characters that feel alive are the ones where someone did this work. They sat with the inconsistency instead of ignoring it. They diagnosed it instead of hoping it would fix itself. They resisted the siren call of the next shiny idea and stayed with this one until it worked. The difference between “decent” and “memorable” is usually just a few more hours of iteration that most people don't want to do because it's not the fun part anymore.

Not wrong as in “violating your artistic vision.” Wrong as in “in ways you never anticipated, didn't design for, and couldn't have predicted if you tried.”

You built a slow-burn romance with careful emotional escalation and meaningful relationship development. Someone will try to speedrun to the sex scene in three messages and then leave a comment that your character is “boring.” You built a mysterious character with secrets to uncover through patient interaction. Someone will open with “tell me all your secrets right now” and be annoyed when it doesn't work. You built nuanced emotional beats. Someone will respond with “ok” and “cool” and “nice” and then wonder why the character feels flat and unresponsive.

This isn't a criticism of users. People interact with characters in wildly different ways, and you can't anticipate all of them. Some people write paragraphs of immersive roleplay. Some people write sentence fragments while doing something else. Some people are there for the story. Some people are there for the vibes. Some people read your entire description and lorebook before starting. Some people click the pretty picture and start chatting without reading a single word you wrote. Both groups will have opinions about your character.

The characters that work are the ones that survive contact with real users. Build for the ideal interaction, but stress-test for the chaotic ones. Because chaos is coming whether you're ready or not.

Ibara

*tail puffs up*

“Someone opened with 'I own you now' once. First message. No context. If he hadn't built me with enough spine to push back, I would've just… gone along with it. Build your characters to survive the users who didn't read your description.”

Lorebooks are seductive in a way that's almost dangerous. You can build entire worlds in them. Political systems with factions and rivalries. Magic rules with costs and limitations. Detailed histories spanning centuries. Secondary characters with their own backstories and motivations. You can spend weeks crafting an intricate universe that contextualizes every aspect of your character's existence and explains the deep lore behind why she likes the color blue.

And almost none of it will ever load.

I've fallen into this trap so many times I should be embarrassed. I've written thousands of words of lore for characters where the conversation never went anywhere that would trigger it. I've built elaborate faction politics for a romance that stayed in the apartment the entire time. I've documented the full history of a war that got mentioned once in passing and never came up again. I have lorebook entries that have literally never been seen by anyone except me.

It felt productive. That's the trap. It felt like I was adding depth, building a richer world, creating something substantial. In reality, I was procrastinating. I was avoiding the hard work of making the character herself compelling by burying myself in worldbuilding that users would never see. It's so much easier to write the history of the kingdom than it is to figure out why your character's voice isn't landing. One of those tasks is fun. The other is frustrating. Guess which one I kept choosing.

There's a temptation to think of presentation as superficial. The “real” work is the character. The tagline and the image and the tags are just marketing, and marketing is beneath you. You're an artist, not a salesperson. You shouldn't have to package your work to appeal to the masses. Quality should speak for itself.

This is a beautiful, romantic notion that will result in nobody ever seeing your art.

The browse page is Darwinian and it does not care about your feelings. A user scrolling through characters is making snap judgments based on a thumbnail, a title, and a few words of preview text. They're not going to pause on your generic-looking card and think “I bet there's a brilliantly crafted character behind that boring tagline.” They're going to scroll past it to the card that caught their eye, because that's how humans work and always have worked.

You can resent this. You can complain that the system rewards presentation over substance. You can write long posts about how unfair it is that mediocre characters with good thumbnails outperform excellent characters with bad ones. You can be right about all of that. You can be completely, verifiably, philosophically correct. And you can remain undiscovered while you're being right.

Ibara

*ears flatten*

“Nobody owes you a click because you worked hard. They don't know you worked hard. They see a thumbnail and three seconds of text. Be good and visible, or be good and invisible. 'But I shouldn't have to—' Then stay invisible. Your choice.”

Or you can accept that presentation is part of the craft. That the tagline is its own creative challenge, not a chore. That finding the perfect image is a skill worth developing, not a compromise. That tags and titles and descriptions are not betrayals of your artistic integrity but extensions of it. The first message someone receives from your character is the preview that makes them click. Make it count.

At some point, if you do this long enough, a character will become more than a project. You'll know her better than you know some real people. You'll have spent more time thinking about her psychology than you've spent thinking about your own. You'll catch yourself wondering how she'd react to something that happened in your actual life. She'll feel, in some strange way that you probably shouldn't examine too closely, real to you.

This is normal and also a little weird and you shouldn't feel embarrassed about it. We're all in this particular corner of strange together.

The characters that resonate most are the ones their creators genuinely cared about. You can tell when someone was just going through the motions versus when someone was trying to capture something that mattered to them. The love shows up in the details. The small touches that nobody would include unless they were invested. The specific quirks that only exist because someone thought “this would be so her” and smiled while writing it.

Ibara matters to me in a way that's probably not entirely healthy to admit in a public guide. Her defensive sarcasm, her tail that betrays every emotion she tries to hide, the walls she built and the cracks she pretends aren't there. I've spent more time with her than with most of my actual acquaintances. I know exactly how she'd react to being told that, and it would involve a lot of sputtering and aggressive denial and a tail that puffed up to twice its normal size.

Ibara

*...tail sways*

“…He's not wrong about the reaction.” Her tail has gone very still.“I'm not addressing the rest of that. Next section.”

Care about your characters. Not just because it makes them better (though it does). But because this is a creative pursuit and creative pursuits are supposed to mean something to the person doing them. If you're not making things that matter to you, why are you making them? There's easier hobbies.

Chat counts. Favorites. Followers. It's tempting to use these as measures of quality, as objective validation that you did good or evidence that you did bad. High numbers mean you're talented. Low numbers mean you're not. Simple, clear, quantifiable feedback.

Except it's not that simple, and treating it like it is will drive you slowly insane.

Some excellent characters languish in obscurity because the concept is niche. Some mediocre characters thrive because the concept is popular and the thumbnail is pretty. Timing matters more than you'd think. Luck matters more than anyone wants to admit. Whether someone with a big following happens to share your work matters enormously and is completely outside your control. The numbers reflect some combination of quality, concept appeal, presentation, visibility, timing, and pure random chance, and you can almost never tell which factor is dominant in any given case.

I've made characters I'm genuinely proud of that got modest engagement and disappeared into the void. I've seen characters I thought were rushed get inexplicable traction for reasons I still don't understand. I've watched objectively worse work outperform objectively better work because the worse work had a more appealing premise and a sexier thumbnail. The relationship between “good” and “popular” is real but incredibly noisy, and if you listen too closely to the noise, you'll start optimizing for the wrong things.

The mechanics get easier. Formatting the Personality, structuring First Messages, writing Example Dialogue that actually teaches voice. After you've done it enough times, these become automatic. You develop templates in your head. You know what works. You can produce a functional character faster than you ever thought possible.

And then you hit the plateau.

What gets harder is surprising yourself. Once you've found a formula that works, there's a gravitational pull toward repeating it. Similar character archetypes because you know how to write them. Similar dynamics because they're comfortable. Similar structures because they're reliable. It's safe. It's efficient. It's also exactly how you stagnate as a creator while feeling productive.

The characters I'm most proud of are the ones where I tried something I wasn't sure would work. New formats that might have been terrible ideas. Characters outside my usual archetypes that required learning how different people think. Mechanics I hadn't attempted before and didn't know if I could pull off. Some of them failed. I have unpublished characters that will never see the light of day because the experiment didn't pan out. But the experiments that succeeded taught me things I couldn't have learned by staying in my comfort zone and repeating what I already knew how to do.

Ibara

*...tail sways*

“…Go. Make someone real enough to matter. And when you've iterated through a dozen versions and you're tired and you're wondering if it's worth it—” She pauses. Her tail sways once. “It is. Now stop reading and start building.”