Persona

Who you are, who you talk to, and what to avoid. Grounds every AI feature in the product, alongside the voice.

Studio, then Settings, then the Brand tab.

The fields

Field	Limit	What goes in it
About the business	20,000 characters	Products, stack, real experience (things you built, broke, or learned), and the opinions you actually hold
Content niche	255 characters	What your content is about, as a category. 5 to 10 words
Target audience	1,000 characters	Who buys, or who follows. One sentence
What to avoid: never do this	4,000 characters	Hard rules for every AI output: words, patterns and claims it must never make

Nothing in "About the business" should be something you would not say out loud: comment drafts and posts pull from it directly.

Interview me

Staring at empty boxes is hard, so there is an Interview me button.

It reads whatever you have already filled in, asks you tailored questions, and turns your answers into the fields above. Nothing is saved until you apply it, so you can back out.

Use it for a first pass, then edit by hand.

Examples

About the business

Solo founder building Postlyra, a social media tool for other solo founders.
PHP and Laravel on the backend, React on the front. I ship in public and
write about what breaks.

Content niche

Social media automation for solo founders and small teams.

Target audience

Solo technical founders and indie hackers, mostly 1 to 3 years in, doing
their own marketing and hating it. They can read code. They are allergic to
growth-hacker language and to advice from people who have never shipped.

What to avoid

Never claim revenue or growth numbers I have not verified.
No advice about things I have not done myself.
Nothing about hiring, fundraising, or scaling teams. Not my world.
No engagement bait, no "comment YES if", no fake vulnerability.

Memory

The same Brand tab holds facts (short, reusable statements about the brand: a pricing detail, a policy, a fact worth repeating exactly) and reference documents (longer material an AI generation can pull from). Add facts by hand, or accept ones the product suggests after a post publishes, when "Learn from what I publish" is on.

Every AI call that used a fact or a document records which ones, so a post's "What the AI used" panel can point straight at them. Editing a fact there is the fix for the next post.

Persona versus voice

They pull in different directions on purpose.

Persona, this page, is what to talk about and to whom.

Voice, the other Settings tab, is how it sounds: non-negotiables, structure, a writing sample, and your content pillars. See Voice.

A thin persona gives you well-written posts about nothing in particular. A thin voice gives you relevant posts that sound like a press release. You want both.

Where it is used

Settings shows a "Where it's used" table naming, feature by feature, which parts of the persona and voice it reads. Broadly:

• Every AI writing feature (Write with AI, Rewrite, autopost rewriting, repurposing, script and hook generation)
• The quality check that runs before a draft is approved
• Subreddit and channel suggestions
• Idea mining through the CLI and MCP, which fetch it as part of the mining context
• Inbox reply drafts

When output feels off

Check persona and voice before you blame the model. The usual causes:

• Generic output means the niche is too broad. "I build software" tells it nothing.
• Wrong reader means the audience field is missing what they actually care about.
• Claims you would not make means the avoid field is empty. That field is doing real work.