This guide walks you through the privacy settings in ChatGPT, Claude, Gemini and Microsoft Copilot, how to clean up what you've already shared, and what those switches can't stop. Allow 20 to 30 minutes for all four.
Why bother
A contract, a payslip or a doctor's note pasted into a chatbot goes onto the company's servers. Depending on your settings, it can train future models, feed a long-term "memory" about you, or be read by human reviewers. Google tells Gemini users not to enter confidential information they wouldn't want a reviewer to see. No switch makes a chat disappear, so Step 1 matters most.
Before you start
- Time: about 5 minutes per assistant, plus clean-up time.
- What you need: each account you use, signed in. Desktop or web is easiest, because some settings are hidden on mobile.
- Checked on: 12 October 2026, against each vendor's help pages: ChatGPT web, iOS and Android (help pages updated September 2026), Claude web, desktop and mobile (consumer Free, Pro and Max), Gemini web and app (personal Google accounts), and Microsoft Copilot (the updated app on web, Windows and mobile, plus the older app). Menus move often. If a name doesn't match, look for the closest wording under Settings, Privacy or Data controls.
- Work or school accounts: if you sign in with an employer or school account (ChatGPT Business or Enterprise, Claude Team, Google Workspace, Entra ID), your admin controls these settings and may be able to see your content. This guide covers personal accounts.
Steps
1 Strip identifying details before you paste anything
This works on every assistant, whatever its settings.
- Paste the document into a plain text editor first, not straight into the chat.
- Replace names, personal ID numbers, addresses, account and policy numbers, signatures and dates of birth with placeholders like
[CLIENT]or[DOB]. - For medical notes, keep the facts you need help with and remove who and where: your name, the clinic, the doctor and record numbers.
- Only paste the clause or paragraph you need help with. Uploaded PDFs and photos are handled like typed text.
You should see a version that still makes sense but wouldn't identify anyone to a stranger.
2 ChatGPT: turn off training
- Web or desktop: click your profile icon, then
Settings > Data controls. - iPhone and Android: open the sidebar, tap your profile icon, then
Settings > Data controls. - Select
Improve the model for everyoneand switch it off. On the web, then selectDone.
You should see the option shown as off, across all your signed-in devices. New chats won't be used to train OpenAI models, but they are still stored in your history.