Style learning
How DUGGAI builds a model of your voice — and what you can do to push it in the right direction.
What style learning is
DUGGAI maintains a per-account style profile: the words you use, your average reply length, your formality, your sign-offs, the way you handle different categories of contact. The profile is built from three signals.
The three signals
1. Your sent mail
On connect, DUGGAI samples the last ~30 days of your sent folder to seed the profile. Sent mail is the most reliable signal because it's actually you, in your voice, replying to real people.
2. Your edits
Every time you edit a draft before sending, DUGGAI records the diff. Edits are the strongest signal: if you change "Best regards" to "Cheers" ten times, DUGGAI stops writing "Best regards."
3. Your approve / reject decisions
Approving a draft as-is says "this is on-target." Rejecting one says "not even close — try again." Both feed the profile, but rejections are weighted heavier because they're a clearer signal.
Per-contact style
DUGGAI tracks how you write to each contact specifically. You probably write differently to your boss than to a vendor than to your sister — the model picks this up over time. For very stable patterns (e.g. "always casual with this person"), you can set a hard rule from the contact panel.
Manual style notes
In Settings → Styleyou can add free-text style notes that apply globally. Useful when there's a rule you want to enforce immediately rather than wait for the model to learn:
- "Never use exclamation marks."
- "Always sign off with my full name and title."
- "Keep replies under three sentences unless the question requires detail."
What style learning is not
Your style profile stays inside your account. It's not used to train shared models, not pooled with other users, not exported. See Privacy & data for the details.