Writing in your voice
How AI Emaily writes email that actually sounds like you.
Most AI email tools produce the same polite, generic reply for everyone. AI Emaily doesn't. Here's the exact process — voice learning, client profiles, context variables, rules, and Copilot review — that turns an incoming message into a draft you'd have written yourself.
01—The problem
“Sounds like AI” is a dealbreaker.
A reply that reads like a template costs you the thing email is for: trust. Clients notice the stiff greeting, the corporate hedging, the em-dash-free monotony of a language model's default voice. Send enough of them and you sound like a bot — or like you didn't care enough to write it yourself.
So the bar isn't “grammatically correct.” It's indistinguishable from you. That takes more than a good model. It takes knowing how you write, who you're writing to, the facts that must be exact, and the rules you never break — and then keeping you in the loop before anything is sent. That's the whole pipeline below.
The process
Six steps, every single draft.
From the moment a message lands to the moment a reply is ready, AI Emaily runs the same disciplined pipeline — so the output is consistent, grounded, and unmistakably yours.
Step 01
Learn your voice
From your real sent mail — not a survey.
On connect, AI Emaily builds a private voice model from your own sent messages: how you greet, how long you write, how formal you are, the phrases you lean on. Nothing is invented — it's a mirror of how you already write.
You can steer it directly too: paste a few sample replies, set a baseline tone (e.g. “warm but concise”), or keep separate voices for separate accounts (personal vs support@).
- Trained on your sent folder, per account
- Add sample emails to sharpen it
- Distinct voice per mailbox or persona
Step 02
Add your personal context
The knowledge you add under each mailbox.
You add your own context under each mailbox — who you are, how you work, your offers and pricing, your policies, your boilerplate, anything the AI should always know. Add as many entries as you need and organize them into folders (Bio, Pricing, Policies, Templates — whatever fits how you work).
When you reply from that mailbox, AI Emaily writes directly with the personal context you've added. The draft already knows your details, your positioning, and your standard answers — so you never have to re-explain yourself to it.
- Add unlimited personal-context entries per mailbox
- Organize them into your own folders
- Drafts pull from it automatically — no re-explaining
Step 03
Load the client profile
The business context behind the conversation.
For a client, customer, or lead, AI Emaily loads their profile: the company and their role, the open deal or support ticket, where they are in your pipeline, files you've shared, and any notes you've pinned.
That's what stops a reply from being generic. The draft references the actual deal and the actual thread, because it's grounded in a real profile — not a blank prompt.
- Per-client memory: deal, ticket, and stage
- Company, role, and shared files in scope
- Pinned notes and history pulled automatically
Step 04
Fill in your variables
Facts stay exact — never hallucinated.
Context variables are the hard facts about you and each client: your booking link, your rates, a policy number, an address, a renewal date. AI Emaily inserts these from your saved values instead of guessing them.
If a fact isn't known, the draft leaves a clearly marked blank rather than inventing one — so a wrong figure never goes out under your name.
- Reusable {{variables}} for you and per client
- Deterministic — pulled, not generated
- Missing value → visible placeholder, never a guess
Step 05
Apply your rules
Your guardrails shape every draft.
Rules encode your do's and don'ts: “always offer two time slots,” “never quote a price without CC'ing my manager,” “keep replies to new leads under 80 words,” “sign client mail with the full signature.”
Rules also decide what happens next — which drafts wait for you, which can go out on their own, and which get routed, labeled, or escalated first.
- Tone, length, and content constraints
- Compliance guardrails (what must / must not be said)
- Routing: draft, hold, send, or escalate
Step 06
Draft, review, and learn
You stay in control — and it gets better.
Copilot writes the reply and holds it for your approval. You read it, tweak a word if you want, and send — or reject it. Autopilot can send the routine ones on its own, but only inside the limits you set, and every action is reversible with a full audit trail.
Then the loop closes: every edit you make is a signal. If you always shorten its drafts or swap “Hi” for “Hey,” the voice model adjusts, so next week it needs fewer edits.
- Copilot: you approve every send in v1
- Autopilot: autonomous within your boundaries
- Undo + audit on every action
- Learns from your edits over time
02—The voice model
What “your voice” actually means.
Voice isn't vibes — it's a set of observable signals we learn from your sent mail and mirror in every draft.
Greeting & sign-off
“Hey Sam,” vs “Dear Mr. Patel,” — and how you close: “Best,” “Cheers,” or just your name.
Length & density
Two-line replies or full paragraphs. We match how much you actually write to this kind of person.
Formality & warmth
Buttoned-up with a new client, loose with a teammate. Tone shifts by who you're writing to.
Phrases you reuse
Your go-to lines — “circling back,” “let me know if that works” — surface where they fit.
Rhythm & punctuation
Em dashes, one-line paragraphs, the Oxford comma or not. The small tics that read as you.
Emoji & casing
A 👍 or never. Lowercase-i or full sentence case. We mirror it — we don't impose a house style.
03—Context & client profiles
The same email to two people should read differently.
A generic assistant treats every recipient the same. AI Emaily keeps a profile per person — the running memory of your relationship: their name and role, every past thread, the open deal or ticket, files you've exchanged, and notes you pin.
When you reply, that profile is loaded into the draft. So the message to a first-time lead is warm and explains the basics; the message to a client of three years skips the throat-clearing and picks up exactly where you left off. Same intent, two very different — and correct — emails.
What a profile holds
- Identity — Name, role, company, and how you address them.
- History — Every prior thread, summarized and searchable.
- Variables — Their policy number, booking link, rate, address — exact, reusable.
- Files & notes — Attachments shared and anything you've pinned.
- Stage — Where they are — new lead, active deal, renewal, support.
04—Rules & control
Your rules decide what it says — and who sends it.
Voice makes a draft sound like you. Rules make sure it says the right thing, and control decides whether it waits for you or goes on its own.
What rules can enforce
- Tone & length by recipient type (new lead vs client vs vendor)
- Content guardrails — always include X, never promise Y
- Compliance lines that must appear (or must never)
- Always attach the booking link / signature / disclaimer
- Route, label, or escalate before drafting
Manual
The agent drafts only when you ask. Nothing happens on its own. Full control, zero surprises.
Copilot
Every reply is drafted in your voice and queued for your approval. You read, adjust, and send. The default, and the only sender in v1.
Autopilot
Routine, rule-matched replies go out on their own — inside the caps, allowlists, and guardrails you define. Always reversible, always logged.
The feedback loop. Whatever you change before sending — a shorter reply, a different greeting, a dropped phrase — feeds back into your voice model. The drafts need fewer edits every week. Nothing about this is static.
05—Privacy
Your voice stays yours.
A model of how you write is sensitive. We treat it that way.
Never trained on your mail
Your voice model is private to your account and never used to train shared or third-party models.
Encrypted + BYOK
Sensitive data is envelope-encrypted, and you can bring your own model key so drafting runs on a key you control.
Zero-retention + audited
Zero-retention with model providers, and every AI action is logged and reversible — no black box.
Questions
The honest answers.
No. Your voice model is private to your account and is never used to train shared or third-party models. We run zero-retention with our model providers, everything sensitive is encrypted, and you can bring your own model key (BYOK) so drafting runs on a key you control.
Send email that sounds like you — not like AI.
Start free. Add a card when you're ready.
- No credit card
- Free plan forever
- Every provider