Feature guide
Email voice training: drafts that sound like the sender, not a template
People do not have one universal email tone. A useful voice model should preserve recognizable habits while adapting to the relationship and purpose of each message.
Updated 21 July 2026 · 6 min read
What “your voice” means in email
Voice is a bundle of repeatable choices: sentence length, directness, warmth, greetings, closings, punctuation, structure, and the amount of explanation you usually provide. It is not a gimmicky personality label.
A strong model also notices context. Many people are brief with coworkers, more explanatory with clients, and warmer with long-term contacts. Flattening those differences is one reason AI-written email often feels artificial.
How Gralvis learns a voice profile
With mailbox permission, Gralvis samples sent messages and summarizes observable writing patterns. It can segment patterns by relationship so the most relevant style is selected for a new draft.
The system does not need every message you have ever sent. A representative sample is more valuable than a large pile of signatures, automated notifications, and one-word replies.
Facts and voice are different problems
Sounding like you does not make a reply correct. Product details, prices, guarantees, and policies belong in Business Brain, where you can review and update them directly. Keeping fact grounding separate from voice makes both easier to inspect.
Edits are the most honest quality signal
A thumbs-up can mean “good enough.” The final sent reply shows what you actually wanted. Gralvis measures the normalized edit distance between its generated body and the final body, then stores that feedback for retraining.
A falling edit rate is a practical sign that personalization is improving. A rising rate may point to stale business context, the wrong relationship segment, or a new category of message.
Frequently asked questions
How many sent emails are needed?
Quality depends on variety more than a universal number. Include enough genuine examples to show greetings, typical explanations, and different relationship styles.
Can I change the learned style?
Yes. You can retrain as your writing changes, and edits to drafts provide direct correction signals.
Will every reply use the same tone?
No. Gralvis can select a voice segment that better matches the recipient relationship and thread context.