Which tool to choose
For writing, the normal assistant is enough: you explain to whom and why, and it prepares the draft. For sending automatically you need integrations with email or with messaging apps, which some assistants offer and which must be authorized carefully. The practical advice: use the AI as an author and keep sending in your hands. If you really want to automate sending, do it only for repetitive, harmless messages, and keep control. The safest tool here is you, pressing send after reading.
How to do it
- Explain the message to the AI: recipient, purpose, tone, points to cover.
- Have it write and read the draft carefully: right recipient? no promises or errors?
- Correct what's off and send yourself from your email or messaging program.
- Only for truly repetitive, low-risk messages, consider an integration that sends, while keeping control.
A concrete example
Antonio had to reply to twenty similar emails from clients asking for information. He used the AI to prepare a draft reply for each, tailored to the question. But he didn't connect automatic sending: he read each draft, corrected two wrong details, and pressed send himself. It took him far less time than writing them from scratch, and no email went out with the error that, without rereading, would have reached the client. The AI provided the speed; the safety came from the last act being his.
When it DOESN'T work (and how to fix it)
If the AI gets the recipient or content wrong
It's exactly the risk that makes automatic sending dangerous. With rereading before sending, you intercept and correct it. Without it, the error reaches its destination. That's why, by default, the last check stays human.
If you need to send many similar messages
Have the AI prepare them all as drafts and send them yourself in a batch, reading them quickly. It's almost as fast as automation, but without the risk of blindly sending out twenty messages with the same error multiplied.
If you really want to automate sending
Limit it to controlled cases: standard messages, safe recipients, content that doesn't change. Keep a log of what gets sent and a way to stop everything. Automating sending isn't forbidden, but it must be treated as a delicate action, not as just any convenience.
A tip from someone who really uses it
Sharply separate writing and sending, and give the AI only the first. Writing is where the assistant saves you real time: drafts, tone, structure, adapting to the recipient. Sending is an act with consequences — toward a client, a boss, a friend — and the moment a message goes out is the moment you expose yourself. Keeping your finger on the "send" button slows almost nothing and saves you from errors you can no longer recall. The convenience of automating sending is real, but it rarely outweighs the risk of a wrong message sent out under your name.
Frequently asked questions
Can the AI send emails by itself?
With the right integrations, technically yes, but it's a delicate action. For most uses it's better to stop it at the draft and send yourself: writing saves you time, sending you control.
When does it make sense to automate sending?
Only for repetitive, standard, low-risk messages, with safe recipients and active oversight. For anything personal, delicate, or important, sending stays yours.
How do I reduce the risk of errors?
By always rereading before sending: recipient, tone, content, any promises. It's the check that intercepts the mistake while you can still correct it.
If I connect my email, can I let the AI handle the replies on its own?
No, and it's the shortcut that sooner or later sends the wrong message to the wrong person. The AI doesn't grasp the context like you do: it can reply to a client with another's tone, misunderstand a request, send an error to its destination with full confidence. And a sent email can't be recalled. Keeping sending in your hands isn't excessive distrust, it's the only point where you can still stop a mistake. Delegating to the AI both writing and sending is the quickest way to discover the error when it's already in someone else's inbox.