Setting up your writing voice

The voice calibration is what makes distro useful past the novelty phase. Every AI tool promises personalization and then spits out the same "I'm excited to share…" template. We do three concrete things to avoid that.

1. Writing samples

Paste 1–3 pieces of your writing (20–2000 chars each). Tweets, LinkedIn posts, Slack messages, blog snippets — anything that sounds like you. More samples = better calibration. We analyze them with a fast model (Haiku-class) and extract:

• A 4-axis tone profile (formal ↔ casual, professional ↔ personal, serious ↔ witty, curated ↔ raw)
• 3–5 distinctive attributes ("uses concrete numbers", "self-deprecating asides", etc.)
• A one-line voice summary that every future draft prompt includes

2. Tone sliders

After analysis, you can drag the four sliders to fine-tune. If the samples you pasted were from a serious context but you want repurposed drafts to skew witty, adjust here.

3. Banned phrases

The product ships with a default banned list — "unlock," "leverage," "game-changer," "synergy," "excited to share," "thrilled to announce," and about thirty more. Add your own. These literally never appear in generated drafts; prompts include them as hard constraints.

Per-app overrides

Each app can override the global voice — useful if you run a B2B tool in one workspace and a personal newsletter in another. Settings live under Customize → Voice.

What "calibration" is not

It's not a fine-tune of any model. Your samples are used as context in each prompt, not training data. We never send your writing to provider partners or use it for model training. See the privacy policy for the full story.