Guide
Using AI without worrying about your privacy
Worried about what happens to what you type into an AI chatbot? You have more control than you might think. Here is what the big AI apps really do with your chats, the settings that switch training off, a simple traffic-light guide to what's safe to paste, and a two-minute checklist. Every setting is checked against the company's own help pages.
Myths vs facts
Myth A person reads everything I type.
Fact No. Chats are processed by computers. Some services have people review a small share of chats to improve quality and safety, and Google says reviewed Gemini chats are kept for up to three years, separate from your account. support.google.com
Myth Everything I type trains the AI, and I can't stop it.
Fact Free personal accounts may use chats to improve the AI, but every major assistant has a setting to switch that off, and most offer a temporary or incognito chat.
Myth My company's AI account works the same way as my free one.
Fact Usually not. OpenAI, for example, says it does not train on data from ChatGPT Business, Enterprise or its API by default. Business plans also come with contracts and admin controls. openai.com
Myth Deleting a chat removes it instantly.
Fact Deleted and temporary chats are usually removed within about 30 days. Copies may be kept longer for safety reviews or legal reasons, so the best protection is not to share sensitive details at all. help.openai.com
What each AI app does with your chats, and how to change it
Settings for free and personal accounts. Menus move around, so if you can't find a switch, search the app's help pages for "privacy".
ChatGPT
- By default
- Personal accounts: chats may be used to improve OpenAI's models.
- Turn it off
- Settings → Data controls → turn off "Improve the model for everyone".
- Private option
- Temporary Chat: not saved to your history or used to improve models; OpenAI may keep a copy for up to 30 days for safety. It can still use your saved memories.
Business, Enterprise, Edu and API data is not used for training by default.
Source: help.openai.comClaude
- By default
- Personal accounts choose: allow training and chats are kept for 5 years, or don't and they are kept for 30 days.
- Turn it off
- Settings → Privacy → turn off the option to help improve Claude.
- Private option
- Deleted conversations are not used for future training.
Claude for Work, Education, Government and the API are not covered by consumer training.
Source: anthropic.comGemini
- By default
- "Keep Activity" is on: chats are saved to your account and may be used to improve Google's AI; a small share is reviewed by people.
- Turn it off
- Gemini → Settings → Activity → turn off Keep Activity, or set chats to auto-delete.
- Private option
- Temporary chats and chats with Keep Activity off are kept for 72 hours and are not used to train Google's models.
Chats reviewed by people are kept for up to 3 years, separate from your account.
Source: support.google.comMicrosoft Copilot
- By default
- Most personal accounts: conversations may be used to train Copilot's models.
- Turn it off
- Profile → Account → Privacy → turn off model training on text and voice.
- Private option
- The change applies within 30 days. Work and school accounts are excluded from training.
Opting out of training doesn't stop use for safety, security and some product purposes.
Source: support.microsoft.comMeta AI
- By default
- Uses public posts and comments from adults on Facebook and Instagram to train its AI; private messages are not included.
- Turn it off
- In the EU, use Meta's objection form; elsewhere, keep posts private if you don't want them used.
- Private option
- Don't share anything with Meta AI chats that you wouldn't post publicly.
Meta paused the plan in Europe in 2024 after regulators objected, then resumed it in 2025.
Source: wsls.comWhat's safe to paste? A traffic-light guide
Fine to share
- General questions and explanations
- Public information and news
- Your own drafts with names and details removed
- Recipes, travel ideas, learning a topic
Think first
- Work documents: only in a tool your employer has approved
- Your own health or money questions: leave out names and numbers
- Photos of yourself or your family
- Anything you would not want in a data breach
Never paste
- Passwords, PINs and one-time codes
- ID, passport, bank or card numbers
- Other people's health, financial or personal details
- Client secrets, unreleased plans or source code from work
Want maximum privacy? Run AI on your own computer
These free apps run open AI models entirely on your own machine, so nothing you type leaves it. They need a reasonably powerful computer, and the models are less capable than the big online assistants.
What went wrong, and what we learned
- 2025
Shared ChatGPT chats showed up in Google
More than 4,500 conversations that people had shared with a "make discoverable" option appeared in search results, some with names and work details. OpenAI removed the option.
Lesson: Treat a share link like a public post. searchenginejournal.com
- 2024
Italy fined OpenAI €15 million
Italy's data protection authority found OpenAI had used personal data to train ChatGPT without a proper legal basis and had not reported a 2023 data breach. OpenAI said it would appeal.
Lesson: Privacy laws apply to AI, and regulators enforce them. thejakartapost.com
- 2023
Samsung engineers pasted secret code into ChatGPT
Within a month, staff shared confidential information with ChatGPT three times, including source code pasted in to fix a bug.
Lesson: The biggest risk is usually what we choose to paste. cshub.com
Your rights
In many countries the law gives you control over your data, including what AI companies hold. In the EU and UK, the GDPR lets you ask a company for a copy of your personal data and ask it to delete it. India's Digital Personal Data Protection Act gives similar rights to access, correct and erase your data (MeitY). Most AI apps let you download or delete your data from their privacy settings.
Your 2-minute privacy checklist
- Switch off training. In each AI app you use, turn off the setting that uses your chats to improve the model.
- Use private chats for one-offs. Temporary or incognito chats are best for anything personal.
- Use your work account at work. Company-approved business plans have stronger protections than personal ones.
- Strip the details. Replace names, numbers and addresses with placeholders before pasting.
- Check share links. Delete links you no longer need, and never share chats with personal details.
- Clear old history. Delete chats you don't need, or set them to auto-delete.
How we keep this accurate
AI companies change their privacy settings often; Anthropic, for example, changed its defaults for personal Claude accounts in 2025. Each setting on this page links to the company’s own help page or a reliable report, and we recheck them every quarter. This page explains the general picture and is not legal advice: for work, follow your employer’s rules on which AI tools are approved. For other ways to stay safe, see how to spot AI fakes and scams.