Voice
How you write, as opposed to what you write about. Studio, then Settings, then the Voice tab.
The fields
| Field | Limit | What goes in it |
|---|---|---|
| Non-negotiables | 8,000 characters | Rules every generation must follow: banned words, formatting bans, things that must always be true |
| Structure | 8,000 characters | The shape a post usually takes: hook, body, close |
| More notes | 8,000 characters | Anything else worth saying about how you write |
| Writing sample | 20,000 characters | A real piece of your own writing, pasted in whole |
There is no AI generator for the writing sample. A sample the model wrote teaches it nothing: it has to be yours.
What works
Telling a model what not to do is more effective than telling it to "be authentic". A specific non-negotiables list, with actual banned words and patterns, produces something that sounds like you. A vague one produces the generic AI voice everyone can spot.
Simple, direct, conversational English. Not a native speaker, so no
vocabulary I would not say out loud.
Banned words: delve, leverage, game-changer, unlock, elevate, robust,
seamless, unleash, dive in, in today's world.
Never use an em dash. Use a period, a comma, or parentheses.
First person. This is me talking, not a brand account.
Honest and slightly blunt. If something failed, say it failed. No spin.Refine it over time. When a generated draft sounds wrong, the fix usually belongs here rather than in the draft.
Per-network overrides
A separate box per social network, appended to the voice as its own section when a piece of writing is aimed at that network specifically (the composer's picked channels, autopost, the quality check, comment drafts, and so on). A writer that does not know its network yet, like a hook or a subreddit suggestion, gets the brand voice alone.
Until you save this tab once, every network's box holds that platform's own hard convention, not an opinion about your voice: things like character limits and hashtag habits. Save the tab and only what you wrote applies from then on.
Only networks with a working connector show up here. Add an override for one you have not connected yet with Add network.
How I reply
A separate field for replies specifically (Engage → Inbox, and comment drafts): how you answer someone directly, as opposed to how you write a post. Short by default: answer first, no pitch, no sign-off.
Content pillars
The 3 to 5 themes you post about. Every idea and draft can be tagged with one, which is what makes per-pillar performance analysis possible later.
Each pillar has a title, a description, and its own tone note: how this specific theme should sound, layered on top of the brand voice.
A: build-in-public. What happened this week, plainly. Wins and misses both.
Tone: matter-of-fact, numbers over adjectives.
B: technical takes. Opinionated, backed by something I built or broke.
Tone: confident but never condescending.
C: product value. Talk to the reader's problem, not the feature list.
Tone: warm, concrete, no jargon.Up to a handful of pillars: past that, a post cannot be filed meaningfully. A pillar's key (up to 4 characters) is what a post's pillar field stores.
Where it is used
Settings shows a "Where it's used" table naming, feature by feature, which parts of the voice and persona it reads. See Persona.