Voice

The first question anyone asks about a tool that writes for them is whether it will sound like them. This is how Ballad answers it: not with a slider marked casual, but with the things an editor would actually hold a draft to.

Two voices

A workspace has two speakers, and they're not interchangeable:

• You, the founder. First person, your opinions, your experience. This is the voice that says "we shipped the wrong thing for fourteen months".
• Your company, the brand. It speaks for the product without pretending to be a person.

The plan decides which one should say what, and where. Bring in teammates and each gets their own founder voice; nobody posts as anyone else.

What a voice knows about you

Set up in the Tailor your voice walkthrough and edited any time under Settings → Voices:

Your role	Founder, CTO, Head of Ops — it changes what you can credibly claim.
What you write about	Your beats, one per line.
Who you write for	The reader on the other end.
Terms	Say this, not that. shipped not released; founder not solopreneur.
Distinctions that matter	The differences you'd correct someone on in your field.
Writing samples	A few things you've actually written.

Ballad offers to fill most of it in — it reads your site and your posts for samples, and drafts the rest for you to correct. Correcting is faster than composing, which is the point.

How a draft is held to it

Before a draft ever reaches your Inbox it goes through the same checks an editor would make:

• Your terms. A term on the never list doesn't survive.
• The slop gate. Drafts are scored against the voice first. Generic phrasing, filler and the usual AI tells get rewritten or the draft gets thrown out and written again.
• No repeated openings. Ballad keeps track of how your last pieces began and rotates the opening move, so a feed of your posts doesn't read like a template with the nouns swapped.
• No invented specifics. It won't make up a customer, a number, a screenshot or an anecdote. Specifics come from your interview answers, your product, or research it can point at.
• Your readers' words. What people ask and push back on — in replies, comments and interviews — is in front of it while it writes.

Teaching it

The voice gets better from work you were doing anyway:

1. Edit before approving. Your edit is the correction. The phrasing you changed becomes a sample attributed from an edit, and the next draft starts from it.
2. Say why when you reject. One line. It goes to the analyst, so a recurring reason changes the plan rather than just this piece.
3. Paste something you wrote. Any time, under Settings → Voices. A single good sample moves a voice more than a paragraph of instructions.

Nothing here trains a model that serves anyone else. Your samples and edits shape your workspace's drafts and nothing outside it.

What a voice publishes

The same page holds the operational half: which channels this voice is on, which formats it writes, how often, which accounts it posts from, how much you trust it, and the beat it's on.

Where the brand look fits

Colours, font and logo are one look for the whole workspace, under Settings → Brand — not per voice. A teammate's image post looks like the company while the footer still names who's speaking.

One place the voice steps back

Release notes. A changelog entry is held to what actually shipped, so it skips the voice pass that would otherwise reach for an anecdote. Factual beats characterful when the subject is a version number.

Next

• Content marketing — where depth and specifics come from.
• Inbox — editing and rejecting, which is where teaching happens.
• Settings — voices, brand and the rest.