Lea.sh
All articles GUIDE

How to secure ChatGPT before connecting Gmail

Share
LinkedIn

Want ChatGPT to help with email without handing over your whole private life? Start with these settings and a few limits on what it can do.

There is a good reason to connect Gmail to ChatGPT: searching for a booking confirmation is much nicer than scrolling through three years of email. But the same inbox contains your medical appointments, money problems and password resets. Before you click Allow, spend a few minutes deciding how much help you actually want.

A person offers one envelope to a friendly robot through a violet gate while other personal papers stay close

1. Check which Google account you are connecting

Use the account that contains the information you need. Keep a work inbox separate from a personal one. If the job is finding holiday receipts, there is no reason to add your work documents too.

In ChatGPT settings, open Plugins and choose the relevant app. Some accounts still show Apps during interface changes. Read Google’s permission screen before approving. Select fewer Google apps if you do not need every service offered. What ChatGPT can do depends on the app, your permissions and the controls available to your account.

2. Change the privacy setting people often miss

Open Settings, then Data controls. Turn off Improve the model for everyone if you do not want ordinary new personal chats used for model training. You can keep your chat history. Voice sharing has separate controls, so review those if you use voice.

For a one-off personal question, Temporary Chat can reduce what stays in your history. While temporary, it is not used for model improvement and does not create memories, but OpenAI may keep a copy for up to 30 days for safety. Avoid pasting passwords or sign-in codes even there.

3. Make the first email task a small one

Try asking for a summary of one harmless thread. Then check the original. Did it get the date right? Did it mistake a forwarded message for something you wrote?

If your setup supports sending messages, review the recipient, attachments and full text before approving. An accurate summary can still include something too personal to send to your boss. “Draft a reply” is a useful starting point because you get to read the result before anyone else does.

4. Keep deletion out of your first experiment

Do not start with “clean up everything.” Ask for a proposed list of messages to archive, then check a small batch yourself. Archive, Trash and permanent deletion have very different consequences.

An instruction in a chat is helpful. A permission limit or approval step gives you another opportunity to stop an unwanted action. Use the available built-in controls, especially before an assistant works while you are away.

5. Know how to disconnect before you need to

Return to the connected app’s settings and disconnect the account when you no longer need it. You can also review and remove third-party access from your Google Account. That stops future authorized access through that connection; it does not automatically erase information already received.

Where Leash fits

For supported email activity connected through Leash, you can add message review, checks for private details and pauses before certain destructive changes. Think of it as an antivirus for what your AI does with your accounts. It complements ChatGPT’s settings; it does not watch every conversation or a separate direct Gmail connection.

A good first task is still finding that booking confirmation. Once you trust the result, decide whether the next task deserves any more access.

Sources and further reading