An AI email assistant works in three steps: it connects to your inbox, reads and classifies new messages as they arrive, and drafts a reply in your tone for you to approve before anything goes out. None of that requires you to change how you use Gmail or Outlook — the assistant sits on top of your existing inbox, and it never sends on your behalf.
Step 1: Connecting your inbox
The assistant connects through your provider's own sign-in screen — Google's OAuth for Gmail and Google Workspace, Microsoft's OAuth for Outlook and Microsoft 365. You never hand over your password. The provider issues a token that only covers the specific permissions the assistant asked for, and you can revoke it at any time from your Google or Microsoft account settings, not just from the assistant itself.
This is the same mechanism calendar apps, CRM plugins, and every other tool that "connects" to your inbox already use — nothing proprietary or unusual about it.
Step 2: Reading and classifying new mail
Once connected, the assistant watches for new messages the same way a mail rule would, then runs each one through a classification model that looks at the sender, subject, and body to decide what kind of email it is — a sales enquiry, a support request, an invoice, a newsletter, and so on. Based on that classification it applies a Gmail label or files the message into an Outlook category automatically, so your inbox is pre-sorted before you even open it.
This is what most people mean by "auto-labeling" — it's classification, not magic. The model has seen enough examples of each category to make a confident guess, and it gets more accurate the longer it has learned your specific inbox.
Step 3: Drafting a reply in your tone
For messages that need a reply, the assistant drafts one — but not a generic one. It looks at a sample of your previously sent emails to pick up your vocabulary, sentence length, and typical sign-off, then writes a draft that reads like something you would actually send. If you've added a knowledge base — product docs, a pricing sheet, an FAQ — it pulls relevant facts from there before drafting, so the reply is grounded in your real information instead of a guess.
The draft is saved directly into your Gmail or Outlook drafts folder — not emailed to you, not shown only inside a separate app. It's exactly where you'd already expect a draft to be.
Why you always approve before sending
No draft is ever sent automatically. Every reply an AI email assistant writes sits as a draft until a person reads it, edits it if needed, and clicks send.
This is the part worth being strict about when you evaluate any assistant, not just DraftKite. An assistant that auto-sends is making a judgment call with your name on it, with no chance to catch a wrong price, a misread request, or a tone that doesn't fit the relationship. Keeping a person in the loop is what makes the automation trustworthy enough to actually rely on.
Where the data lives
Because the assistant reads live inbox content, hosting location and compliance matter. DraftKite hosts data in the EU (Frankfurt) and is built around GDPR requirements — connected mailboxes, classifications, and drafts stay on EU infrastructure. If you operate in a regulated industry, confirm this detail for any tool before connecting a real inbox to it.
- Connects via Google or Microsoft OAuth — no shared passwords
- Classifies and labels new mail automatically
- Drafts replies in your tone, grounded in your own knowledge base
- Never sends — every draft waits for your approval
- EU-hosted, built around GDPR requirements
DraftKite runs exactly this pipeline — connect, classify, draft, approve — on your own inbox. Start the 7-day free trial, no card required, and see your first labeled inbox and drafted replies today.