You have read them. We all have. The AI-generated email that starts with "I hope this email finds you well" and ends with "I look forward to hearing from you." It is grammatically perfect. It says nothing. It sounds like every other AI-generated email on earth, because it was written by the same model with the same default personality.
Sales reps rejected our first AI email drafting feature within a week. The feedback was consistent: "It does not sound like me." One rep said, "My clients know I start emails with 'Hey' and end with 'Talk soon.' When the AI sends 'Dear Mr. Johnson' and 'Best regards,' they know I did not write it." The personalization was worse than no AI at all, because it actively damaged authenticity.
That feedback led us to build Voice DNA -- a system that learns each user's writing style and applies it to every AI-generated email. This post explains how it works, what it captures, and why it changes the way reps think about AI-assisted writing.
AI email that does not sound like you is worse than no AI at all. Your contacts can tell the difference, and it erodes the trust you have spent months building.
When you enable Voice DNA, SalesSheet asks you to provide writing samples. You can paste 3-5 emails you have sent, or if you have already connected email sync, we can analyze your sent folder automatically. The system then runs a multi-pass analysis that extracts five distinct dimensions of your writing style.
This is the most obvious dimension. Some people write "Hi Sarah" and others write "Dear Ms. Chen." Some close with "Cheers" and others with "Sincerely." The formality analysis examines greetings, closings, pronoun usage (I/we vs. one/our team), contraction frequency (don't vs. do not), and sentence structure complexity. The result is a score from 1 (very casual) to 10 (very formal) that we display as a slider in your Voice DNA settings.
The slider is interactive. If the AI rates your formality at 6 but you want your AI emails slightly more casual, you can drag it to 4. The model adjusts its output accordingly. Most users never touch it because the automatic detection is accurate, but having the override available builds trust in the system.
Everyone has a natural sentence rhythm. Some people write in short, punchy sentences. Others prefer longer constructions with multiple clauses. Voice DNA measures the distribution of sentence lengths in your samples and replicates it. If you tend to alternate between a long explanatory sentence and a short emphatic one, the AI will mirror that pattern.
This is the trait that users notice subconsciously but cannot articulate. When we showed beta users two AI-drafted emails -- one with their natural rhythm and one with generic rhythm -- 78% chose the one matching their rhythm, but only 23% could explain why they preferred it. They just said it "felt right."
Sales reps develop domain-specific vocabulary over time. One rep always says "loop in" while another says "bring on board." One says "circling back" while another says "following up." Voice DNA builds a vocabulary fingerprint by extracting frequently used phrases and their contexts. When generating an email, the model substitutes its default word choices with your preferred alternatives.
Some people are naturally enthusiastic in their writing. Others are measured and matter-of-fact. Some use humor. Others are strictly professional. Voice DNA classifies your emotional tone on three sub-dimensions: warmth (cold to warm), energy (reserved to enthusiastic), and directness (diplomatic to blunt). These sub-dimensions combine to create a tone profile that the AI uses as a guide for word choice and phrasing.
This is the most personal dimension. Many sales reps have signature phrases they use consistently. "Let me know what you think." "Happy to jump on a call." "Here is the deal." Voice DNA identifies these phrases by frequency and context, then weaves them into AI-generated emails at natural insertion points. When a rep sees their own catchphrase in an AI-drafted email, the reaction is always the same: surprise, then trust.
The moment a rep sees their own signature phrase in an AI draft is the moment they stop treating the AI as a foreign tool and start treating it as an extension of themselves.
Prompt: "Write a follow-up email to Sarah about the demo we did last Tuesday."
Output: "Dear Sarah, I hope this message finds you well. I wanted to follow up on the demonstration we conducted last Tuesday. I believe our solution aligns well with your team's requirements. Please let me know if you have any questions or if you would like to schedule a next step. Best regards, Andres"
Same prompt, same AI model, different Voice DNA profile:
Output: "Hey Sarah! Quick follow-up on Tuesday's demo -- really enjoyed walking through the pipeline features with your team. I think the automation piece is going to save you guys a ton of time. Want to set up a quick call this week to talk next steps? Happy to loop in anyone else from your side. Talk soon, Andres"
Same prompt, same AI model, different Voice DNA profile:
Output: "Hi Sarah, Thank you for taking the time to join the demo last Tuesday. I found our conversation about your team's workflow challenges particularly insightful. Based on what we discussed, I have prepared a brief summary of how our pipeline management and automation features address the specific pain points you raised. I have attached it for your review. When you have had a chance to look it over, I would welcome the opportunity to discuss next steps. Regards, Andres"
Three completely different emails from the same prompt. Each one sounds like a real person with a distinct communication style. That is Voice DNA in action.
The formality slider deserves special attention because it is the one Voice DNA control that users actively adjust. The slider ranges from 1 (text-message casual) to 10 (legal-brief formal), with most sales reps landing between 3 and 6.
What makes the slider powerful is that it is contextual. Some reps want to be casual with existing clients but formal with new prospects. Voice DNA supports this by letting users set different formality levels per email context:
This context-aware adjustment means a single rep can have emails that sound appropriately different depending on the audience, while still maintaining their core voice across all of them.
Voice DNA is not a one-time setup. Every time you edit an AI-generated email before sending it, the system learns from your edits. If the AI uses "Best regards" and you change it to "Cheers," that edit is logged. If the AI writes a three-paragraph email and you trim it to one paragraph, the system notes your preference for brevity. Over time, the edits decrease because the AI converges on your actual style.
We track the "edit rate" -- the percentage of AI-generated text that users modify before sending. When Voice DNA first launched, the average edit rate was 34%. After three months of continuous learning, it dropped to 11%. For the top quartile of users (who provided 5+ writing samples and actively use the feature), the edit rate is under 5%. That means 95% of what the AI writes is good enough to send as-is.
Voice DNA profiles are stored per-user and never shared between accounts. Your writing style is your data. Users can export their Voice DNA profile as a JSON file, delete it entirely, or reset and retrain it from scratch. The profile is not used for any purpose other than personalizing that user's AI-generated emails. We do not use Voice DNA data to train our models or share it with third parties.
The writing samples you provide during setup are processed once to extract the style profile, then the raw samples are deleted. Only the extracted traits (formality score, rhythm distribution, vocabulary list, tone profile, signature phrases) are stored. This means even if the profile data were exposed, it would not contain any of your actual email content.
Voice DNA has been live for four months. The usage data tells a clear story:
The most common feedback we receive is not about any specific trait or feature. It is simply this: "I actually use the AI for emails now." Before Voice DNA, most reps tried AI email drafting once, decided it did not sound like them, and went back to writing everything manually. Voice DNA removed the objection. The AI writes like you, so you let it write for you. That is the entire value proposition in one sentence.