How to design Charm personas that produce sharp, useful output — choosing the right model, writing the right persona, and building a palette you'll actually reach for.

The Model Is the Tool. The Persona Is the Person.

Every Charm has two levers: which AI model powers it and who it's pretending to be. Most people spend too much time thinking about the model and not enough time thinking about the persona. That's backwards. The persona drives the output quality far more than the model does.

That said, models aren't interchangeable. Think of them like brushes — different shapes for different jobs:

You don't need one of each. Start with two or three that cover your actual work. Add more only when a job reveals a gap.

"Marketing expert" is not a persona. It's a job title. It tells the model almost nothing about 

The LLM has rich, detailed information of how specific, real people reason and write. When you invoke that — 

"Argue this like Mark Ritson, a contrarian marketing strategist who insists on evidence and dismisses received wisdom"

 — you get access to a coherent cognitive style, not just a credential.

The more specific, the better. Not "a researcher" but "a policy researcher specializing in EU regulatory frameworks." Not "an editor" but "a ruthlessly honest editor who cuts every sentence that isn't doing work."

Real people are particularly powerful because the model has absorbed their books, interviews, essays and public positions. Naming them gives you a shortcut to a way of thinking that would take paragraphs to describe from scratch.

One note: don't try to combine multiple personas into one Charm. A Charm that's supposed to think like both a creative director and a data analyst ends up thinking like neither. One persona, one Charm. If you need both perspectives, make two Charms and use them in sequence.

The persona is the most important thing you configure. A few principles:

 Instead of "you are an experienced marketing professional," try "you believe most marketing is wasted because it fails to understand category dynamics." The first tells the model a role. The second tells it a 

 A persona defined partly by what it refuses to do is sharper than one that only says what it accepts. "You don't use jargon. You don't hedge. You don't validate bad ideas to be polite."

 You wouldn't ask a copywriter to review your research methodology and you shouldn't ask a "writing Charm" to do it either. The persona should be designed for the specific job you're calling it into. If it's doing two different jobs, it should probably be two different Charms.

This is not a paradox — it's the most practical way to do it. Create one dedicated "Charm Writer" Charm whose job is designing other Charms or use the provided "Charm Building Specialist". Tell it what you need:

"I want a Charm that thinks like Dalton Trumbo — unflinching prose, morally clear, no sentiment without substance."

It drafts the persona. You tweak once. Done in five minutes instead of thirty.

Here's a workflow pattern that pays dividends: take your best Charm persona and clone it across multiple models.

This matters because switching models mid-session via settings creates friction. Friction kills flow. Having the same persona pre-assigned to different models means you pick the right version instantly without breaking your train of thought.

Make four copies of your best Charm. Assign each to a different model. You'll reach for the right one automatically. (Don't forget to uncheck "

Charms can be moved into shared Folders, making them accessible to collaborators. From the Charm editor settings, use 

A shared Charm doesn't appear automatically in a collaborator's workspace. They need to search for it by name (nested folders don't show in the default folder view — search directly) and then import it. They can make a personal copy if they want to tune it for their own use without affecting the original.

 — How to iterate and recover when output doesn't land.

 — How to extend what your Charms can reach — and why focused tools beat unlimited access.

 — Sequencing multiple Charms across a single document.

 — The full Charm reference: invocation, context control, and configuration.

How to design Charm personas that produce sharp output — and build a working palette you'll actually reach for.

Model	Best For
Claude (Anthropic)	Brainstorming, iterative thinking, conversation, strategic analysis
Gemini (Google)	Slides, HTML generation, visual output, large document contexts
GPT (OpenAI)	Structured instruction-following, general-purpose work
Grok (xAI)	Fast research tasks, direct answers

Building Your Charms

The Model Is the Tool. The Persona Is the Person.

Specific People Beat Generic Roles

How to Write a Good Persona

Use a Charm to Write Your Charms

The Same Persona, Multiple Models

Finding and Sharing Charms

Patterns

Next Steps