You open an AI chatbot to help write an email. Then you paste the whole conversation, including your address, your child’s name, and somebody else’s phone number. The chatbot only needed to know that you wanted a polite reply.
AI can help you organize information, explain a confusing paragraph, or turn rough notes into a message. Before you hit send, take a moment to decide how much of your information the tool actually needs.
- Remove passwords, verification codes, and account numbers.
- Replace names and identifying details with placeholders.
- Check screenshots and attachments before uploading.
- Review the tool’s data settings and privacy notice.
- Check what connected apps can read or change.
- Review the whole conversation before sharing it.
1. Remove anything that can unlock an account
Leave out passwords, one-time verification codes, recovery codes, and security answers. If you need help with a login error, describe what happened and copy the error message after checking it for personal details.
The same habit applies to bank account numbers, payment card details, Social Security numbers, and pictures of identification documents. Ask a general question or use a placeholder when the actual number has no bearing on the answer.
2. Give the situation without the full identity
A chatbot can draft a message to a teacher with “[Teacher]” and “[Child]” in the text. You can fill in the names after copying the draft into your email.
| What you want help with | What to share |
|---|---|
| Writing a school email | The reason for the message, the tone, and the action you want. |
| Understanding a bill | The charge descriptions and relevant amounts, with identifying numbers removed. |
| Improving a résumé | Your experience and skills, using placeholders for contact information. |
| Planning a family week | Labels such as “Adult A” and “Child B,” with the times and travel estimates needed. |
Removing a name is only the first step. A workplace, exact address, job title, and detailed story can identify someone together. Read the prompt once more and remove details that don’t help with the task.
3. Look around the edges of screenshots
Before uploading a screenshot, inspect browser tabs, notifications, email addresses, account balances, QR codes, and the header and footer of the document. Crop to the part you need help understanding.
If you redact a file, use a tool that permanently removes the underlying information. A movable black rectangle placed over text can leave that text in the document. Export the edited copy and check it before uploading. For a short question, typing the relevant sentence yourself may be easier.
4. Check how the service handles your conversations
Open the chatbot’s settings and privacy notice. Look for explanations of how conversations are stored, whether they may be used to improve models, when people may review them, and what happens after deletion.
Training controls, chat history, memory, and data retention can have separate settings. Read what each control does. A temporary or private chat mode may still involve some retention or safety review under the service’s policy.
For example, Google’s Gemini privacy guidance explains that some data undergoes human review and advises users to avoid entering confidential information they wouldn’t want reviewed or used to improve its services. Check the current notice for the service and account you’re using.
Read the Gemini Apps Privacy Hub.
5. Review access before connecting email, files, or calendars
A connected app can give an assistant access to information beyond what you type into the chat. Before approving access, check which account you’re connecting, which information it can read, and whether it can create or change anything.
If your goal is to organize three appointments, you could start by providing those three appointments with names and addresses removed. If you choose to connect an account, learn where to review and revoke that access later.
See Google’s explanation of Gemini Connected Apps and data handling.
6. Pause before sharing someone else’s information
A school email, customer message, or family group chat can contain information about several people. Copy only the section needed and remove their identifying details.
For work documents, follow your employer’s rules and use the approved tool and account. A personal subscription and a workplace service may have different data protections. Check the terms that apply to your actual setup before uploading confidential material.
7. Read the conversation before sharing a link
A chat can start with a harmless recipe question and end with details about your household. Before creating a share link, inspect everything the link will expose. Often, copying the finished answer into a fresh document gives you more control over what you send.
Also check generated drafts before publishing them. If you supplied private details earlier in the conversation, look for those details in the answer.
A prompt you can copy
Help me write a polite email requesting a schedule change. Use [Name], [Date], and [Location] as placeholders. Keep the message under 150 words. Ask only for details needed to write the message.
Starting with placeholders makes the final check easier. You can add the real details in your own email or document once you’re happy with the wording.
If you already shared something sensitive
Stop adding more details. Use the service’s deletion and privacy controls, review any share links, and contact its support team if you need help understanding what can be removed. Deletion timing and exceptions depend on the service.
If you exposed a password or recovery code, replace it through the account provider. If the information belongs to your employer or a customer, follow your organization’s reporting process promptly.
Before pressing send, ask yourself, “Which details does this answer actually need?” Remove the rest, then submit the prompt.
Keep reading
- What AI Actually Is, Explained With Things You Already Know
- The Productivity Apps Worth Keeping When Your Digital Life Is a Mess
- Explore our family calendar guide
Further reading — NCSC guidance on large language models and FTC guidance on AI privacy and confidentiality commitments.

