Jump to a section
If you wouldn't write it on a postcard, don't type it into a chatbot. Passwords, codes, bank details, ID numbers and other people's private information stay out. For everything else, swap real details for placeholders and you'll get the same help.
Why your chats aren't completely private
An AI chatbot is a service run by a company. What you type is sent to that company's computers, and depending on your settings it can be kept, used to improve future versions, and in some cases read by people.
Google is unusually direct about this. Its Gemini privacy page says a subset of chats is reviewed by human reviewers, that reviewed chats are kept for up to three years, and it asks you not to enter confidential information you wouldn't want a reviewer to see.
Anthropic lets people choose whether their Claude chats help improve its models. Turning that choice off does not delete a chat. If you allow training, de-identified data can be kept for up to five years; ordinary deletion from its systems within 30 days starts after you delete a conversation.
Each company's rules are different, and they change. The safest habit is the same everywhere: treat a chatbot like a helpful stranger, not a locked diary.
AI chatbots and your privacy: what they keep, who sees it, and how to take control
Is ChatGPT safe? An honest answer, and five settings to check
How to get the same help without the private bits
You can almost always get exactly the same help by describing your situation instead of pasting the real thing. The chatbot doesn't need your name or your numbers to write a good letter or explain a bill.
- Swap names for roles: write 'my landlord' instead of a name, and 'my bank' instead of the bank's name if it doesn't matter.
- Swap numbers for placeholders such as [ACCOUNT NUMBER] or [DATE]. You can fill them in yourself afterwards.
- Describe rather than paste. 'I was charged twice for the same order' works as well as a copy of your statement.
- If you share a screenshot, crop or cover anything personal first.
- Before you send a long message, read it through once and ask: would I mind a stranger reading this?
I've been charged twice for the same order by [COMPANY]. The order was on [DATE] for [AMOUNT]. Write a short, polite email asking for a refund of the duplicate charge. Leave the bits in square brackets for me to fill in.Choose a setting, then keep private details out
In ChatGPT, use Settings, Data controls to review Improve the model for everyone. In Gemini, review Keep activity. Those choices affect model-improvement uses of future conversations, but neither turns a personal chatbot into a confidential advice service.
The detailed guide below covers the feedback exceptions, temporary chats and other assistants in one place. Check that guide when you need the current steps, rather than assuming a menu name will stay the same.
How to stop AI chatbots using your chats for training
AI chatbots and your privacy: what they keep, who sees it, and how to take control
Choose your next step
Pick the guide for the tool or task you use.
How to hide AI Overviews in Google Search
How to turn off Gemini and smart features in Gmail
How to limit Meta AI in WhatsApp
Can you turn off Meta AI on Instagram and Facebook?
How to remove Copilot and turn off Recall in Windows
How to turn off Apple AI features on your iPhone
Is Gemini safe? Privacy checks for a personal Google account
Never type these into an AI chatbot
- Passwords, PINs and one-time security codes
- Bank, card or account numbers
- Passport, driving licence, National Insurance or Social Security numbers
- Someone else's private details, such as their health, address or messages
- Anything your employer, school or anyone else has asked you to keep confidential
Sources
These sources support the guide. Tool settings, prices and policies change, so dated source boxes show when those details were checked.
- OpenAI: data controls in ChatGPT
- Google: Gemini Apps Privacy Hub, updated 26 October 2026
- Anthropic: How long do you store my data?
AI helps with research and drafting. Dan checks the sources and is responsible for what is published.
How we check