Every AI writing tool on the market produces the same output. You ask it to draft a follow-up email and you get something that starts with "I hope this email finds you well" followed by three paragraphs of corporate filler that no human actually talks like. The grammar is perfect. The tone is sterile. Your contacts can tell immediately that a robot wrote it, and they respond accordingly - by ignoring it.
This is a real problem for sales teams. The entire point of email outreach is to start a conversation that feels personal. When every AI-drafted message sounds like it came from the same bland template engine, you lose the one thing that makes outreach work: your voice. The phrases you naturally use, the level of formality you default to, the way you sign off - all of it gets flattened into generic corporate-speak.
We built Voice DNA to solve this. Instead of generating text from a one-size-fits-all model, Voice DNA learns how you actually write by analyzing your real sent messages. The result is AI drafts that sound like you sat down and wrote them yourself - because the AI is mimicking patterns it extracted from messages you actually sent.
Voice DNA scans your sent emails from Gmail and your messages from Slack. It does not just count word frequencies - it extracts patterns across several dimensions that together form a comprehensive communication fingerprint:
The analysis runs through the Claude API, which processes your writing samples and extracts a structured style guide. This is not simple keyword matching - the model understands context, detecting whether you tend to ask questions or make statements, whether you use bullet points or flowing prose, and whether your emails are three sentences or three paragraphs. The output is a JSON-structured profile that captures your writing personality in a format the AI can use during drafting.
Voice DNA connects to your communication platforms through standard OAuth flows. For Gmail, we request read-only access to your sent mail folder - we never read your inbox, drafts, or anything else. For Slack, we pull messages you have sent in channels and DMs. Both integrations use the same OAuth infrastructure that powers our email sync and Slack integration, so if you have already connected those, Voice DNA can start analyzing immediately.
The OAuth scopes are deliberately narrow. We request gmail.readonly for Gmail and users:read plus message history for Slack. We do not request write access to either platform because Voice DNA only needs to read your past messages - it never sends anything on your behalf. This matters for enterprise users who need to justify every OAuth permission to their IT team. The fewer scopes we request, the easier the approval process.
Once connected, the analysis happens in a Supabase Edge Function that fetches your messages, sends them to Claude for pattern extraction, and stores the resulting style profile in your account. The entire process runs in under a minute for most users, and you can see the progress in real time as the analysis completes each dimension of your profile.
Here is the real difference. Without Voice DNA, the AI drafts a follow-up email that reads like a template - correct but lifeless. With Voice DNA, it drafts something that sounds like you sat down and wrote it yourself. The sentence length matches yours. The greeting matches yours. Even the sign-off matches the way you typically close emails.
We tested this internally by having team members read pairs of emails - one written by the person and one generated by Voice DNA - and asking them to identify which was which. The accuracy was barely above chance. That is the benchmark we are aiming for: AI drafts that are indistinguishable from the real thing, even to people who know you well.
The improvement is especially dramatic for people with distinctive writing styles. If you are someone who writes short, punchy sentences and signs off with "Best," the generic AI would give you flowing paragraphs ending in "Warm regards." Voice DNA catches those patterns and reproduces them faithfully.
The technical flow has four stages: collect writing samples, extract communication patterns via AI analysis, generate a style guide prompt, and inject that prompt into every AI drafting context. The style guide acts as a persistent personality layer that shapes how the AI composes content across all AI features - email drafts, chat responses, and even suggested meeting agendas.
The style guide itself is a structured prompt that gets prepended to every AI generation request. It includes instructions like "use short sentences averaging 12 words," "open emails with a direct statement rather than a greeting," and "include the phrase 'let me know' when requesting a response." These instructions are specific enough to shape the output but flexible enough to adapt to different contexts - a follow-up email will still read differently from a cold outreach, but both will sound like you.
On Pro, you get 50 Gmail threads analyzed. On Ultra, 500 Gmail threads plus 250 Slack messages. More samples means a richer, more accurate profile. We found that 50 threads is enough to capture basic patterns - tone, formality, and greeting style. But 500 threads plus Slack messages captures the subtler patterns: how your writing changes between prospects and existing clients, how you adjust formality based on seniority, and which industry-specific terms you use consistently. The difference between a good Voice DNA profile and a great one is sample size.
Privacy was a non-negotiable design constraint from the beginning. Your writing samples are processed by the AI to extract patterns, and then only the extracted patterns are stored - not the raw emails. We never persist your actual message content in our database. The style profile that gets saved is an abstraction: traits, scores, phrase lists, and formatting preferences. You could not reconstruct the original emails from the profile any more than you could reconstruct a face from a description of its features.
The style profile stays in your account and is never shared with other users, even within your organization. Each team member's Voice DNA is completely private. It is never used to train AI models, never aggregated for analytics, and never sold to third parties. You control what gets analyzed and can re-run the analysis at any time, which replaces the previous profile entirely. You can also delete your Voice DNA profile at any time, which removes the style guide from all future AI generations.
This approach aligns with our broader philosophy on AI security: the AI should work for you without requiring you to trust us with your raw data. Voice DNA proves that you can personalize AI output aggressively without compromising user privacy - you just have to be thoughtful about what you store and what you discard.
Voice DNA turns AI from a generic writing tool into your personal ghostwriter. Every draft matches your natural communication style - because it learned that style from your actual messages.