DraftKite never sends an email on its own. Every reply the AI writes is saved as a draft and waits for a person to read it, edit it if needed, and click send. That single rule — approve before send, always — is the core safety principle behind the whole product. It keeps a human responsible for anything that goes out with your name on it, and it turns AI from a risk into an assistant you can genuinely rely on.
Why auto-send is a risk worth avoiding
An AI model is very good at producing a fluent, plausible-sounding reply in seconds. It is not good at knowing when it's wrong. A language model can misread a request, invent a detail that sounds right but isn't, or strike a tone that doesn't fit the relationship — and it does all of that with the same confidence it uses when it's correct. When a tool sends automatically, there's no moment between "the AI decided" and "the customer received it." The mistake is already out the door.
That gap matters most precisely where email matters most: a price quote, a question about a contract, a complaint from an unhappy customer, a note to your boss. These are the messages where a wrong word costs something real, and they're exactly the ones you'd never want a machine to send unsupervised.
The mistakes only a human catches
Most of what can go wrong with an AI-written email falls into a few predictable categories. None of them are exotic — they're the ordinary failure modes of any system that guesses. What they have in common is that a person reading the draft for two seconds catches every one of them, and an auto-send tool catches none.
| What can go wrong | Why the AI misses it | What a quick read catches |
|---|---|---|
| Wrong tone | The model can't feel the relationship — it doesn't know this client is furious or this colleague is a friend | You notice instantly that the draft is too stiff, too casual, or too cheerful |
| Invented facts | The model fills gaps with plausible-sounding detail — a date, a price, a policy that was never true | You know your real prices and policies and spot the one that's off |
| Wrong recipient or context | The model answers the message in front of it without the wider thread or history you carry in your head | You remember what was already agreed and reframe before it goes out |
| Something confidential | The model has no sense of what shouldn't be said in writing | You quietly delete the sentence that shouldn't leave the building |
How approve-before-send keeps you in control
DraftKite drafts a reply and saves it straight into your Gmail drafts folder or Outlook drafts — the same place your own half-written emails live. Nothing is emailed to a separate app, and nothing is queued to send later. The draft simply sits there until you open it.
From there you do what you'd do with any draft: read it, change a line, delete a paragraph, or throw it away entirely. When it's right, you press send — the provider's own send button, from your own account. DraftKite never holds a "send" permission that fires on its own, so there's no setting, no automation, and no edge case where a reply leaves your mailbox without you clicking it.
Approve-before-send isn't a mode you switch on. It's the only way DraftKite works — there is no auto-send toggle to find, forget, or turn on by accident.
What to look for in any AI email tool
This principle is worth checking for in any assistant you consider, not just this one. Before you connect a real inbox, look for a few concrete guarantees:
- Drafts only — the tool writes to your drafts folder and never has permission to send on its own.
- You see the full reply before it goes — no "summary" that hides the actual wording.
- Easy to edit — the draft lands somewhere you can change it, not a locked preview.
- Clear data handling — you know where your inbox content is stored and processed (DraftKite keeps it encrypted and hosted in the EU).
- Reversible access — you can disconnect the tool from your Google or Microsoft account at any time, not just from inside the app.
Human in the loop is the whole point
It's tempting to think the goal of an email assistant is to remove you from the loop entirely. It isn't. The goal is to remove the blank page, the first-draft effort, and the sorting — while leaving the judgment with you. You still decide what's right, what's true, and what tone fits. The AI just gets you 90% of the way there in seconds, so the last 10% is a quick read instead of a cold start.
That's the trade that makes AI email trustworthy enough to rely on every day. DraftKite classifies your inbox, drafts replies in your tone, and then stops — waiting for you. Start the 7-day free trial, no card required, and see how much a draft-and-approve inbox does without ever sending on its own.