Skip to content

Practical security for real people & small teams

What ChatGPT, Gemini, Claude and Copilot do with your chats, and how to opt out

Guide type
Settings guide + checklist
Risk if skipped
Medium
Time
10 min for all four
Cost
Free
Works on
Web, phone & desktop apps
Last verified
2026-10-05

The short answer

  • On personal accounts, all four can use your chats to improve their AI unless you switch it off. Each has a single setting for it, listed below.
  • Turning training off does not mean "nothing is kept". Each company still stores chats for a while for safety, and some keep chats that a human reviewed for years.
  • For anything sensitive, use the temporary or incognito chat mode. It stays out of your history and is not used for training.
  • Whatever the settings say: never paste passwords, one-time codes, card or ID numbers, or other people's private details into a chatbot.

The four chatbots compared

Trains on your chats by default?Setting to turn it offPrivate modeNotes
ChatGPTYes (personal plans)Improve the model for everyoneTemporary chat, kept up to 30 days for safetyDeleted chats are removed from OpenAI's systems within 30 days
GeminiYes, while Keep Activity is on (on by default for adults)Keep ActivityTemporary chat, kept 72 hoursChats reviewed by humans are kept up to 3 years, even if you delete them
ClaudeYour choice: you are asked when you sign upHelp improve our AI models (in Privacy settings)Incognito chat, kept 30 daysIf allowed, chats can be kept de-identified for up to 5 years
Copilot (Microsoft, personal)Yes, unless you opt outTraining on conversation activity, and Training on voice conversationsNone listed in Microsoft's privacy controls; delete chats you don't want keptOpting out does not stop use for advertising, safety or "general product improvements"

Work and school accounts are different: OpenAI does not train on ChatGPT Business, Enterprise or Edu content by default, Google says Gemini with a work or school account may follow different data terms, and Claude's commercial plans are covered by separate terms. If your employer provides the account, ask your IT team what applies.

Turn it off, service by service

  1. ChatGPT

    On the web: open your account menu › Settings › Data controls › turn off Improve the model for everyone › Done. On iPhone or Android: open the sidebar, tap your profile icon › Data controls › turn it off. The setting follows your account to every device.

    Two catches from OpenAI's own help pages: your old chats stay in your history (delete them separately if you want them gone), and if you press thumbs up or down on an answer, the whole conversation may be used for training even after you opt out. For a private question, open a new chat and select Temporary.

  2. Gemini

    Go to gemini.google.com › Settings & help › Activity › Turn off (or Turn off and delete activity). You can also go straight to myactivity.google.com/product/gemini.

    With Keep Activity off, Google says your future chats won't be used to train its AI models and won't appear in your activity, but are still kept with your account for up to 72 hours. While it is on, a subset of chats is read by human reviewers, and Google asks you not to enter anything you wouldn't want a reviewer to see. If you leave it on, at least shorten Auto-delete from the default 18 months to 3 months on the same page. For one-off private questions, click Temporary chat next to New chat.

  3. Claude

    Click your name › Settings › Privacy (or go to claude.ai/settings/data-privacy-controls) and turn off Help improve our AI models. On the phone app it is the same path: your name › Settings › Privacy.

    Anthropic says that once it is off, your previous and new chats are not used in future training runs, though models already trained on them are not changed. For a private conversation, start a new chat and click the ghost icon in the top right to make it an incognito chat. Thumbs up or down feedback is kept for five years, so skip it on sensitive chats.

  4. Microsoft Copilot

    On copilot.com: select your profile icon, then your profile name › Privacy. In the Windows or Mac app: profile icon › Settings › Privacy. On the phone: menu › profile icon › Account › Privacy. Turn off both Training on conversation activity and Training on voice conversations.

    Microsoft notes that this stops training on your future conversations only; they can still be used for advertising, safety and other product improvements described in its privacy statement.

  5. Check what each one remembers about you

    ChatGPT, Gemini, Claude and Copilot can all remember details across conversations to personalize answers. Memory is separate from training: turning training off does not erase memories. Open the memory or personalization section in each app's settings, read what it has saved, and delete anything you would not want quoted back to you, or to someone else using your device.

Your AI privacy checklist

Tick each item as you finish it. Your progress is saved on this device only. These settings change often, so the checklist suggests a date to look again.

0 of 6 done
  • ChatGPT: Improve the model for everyone is off

    Settings › Data controls. Skip if you do not use ChatGPT.

  • Gemini: Keep Activity is off, or auto-delete is 3 months

    Settings & help › Activity.

  • Claude: Help improve our AI models is off

    Settings › Privacy.

  • Copilot: both training switches are off

    Profile › Privacy: conversation activity and voice conversations.

  • Saved memories reviewed and cleaned

    Delete anything about your health, money, family or work you would not want repeated.

  • Old chats with personal details deleted

    Search your history for addresses, account numbers or names and delete those chats.

What never to type into a chatbot

Don't pasteWhyDo this instead
Passwords, recovery codes, one-time codesAnyone who gets into the account, or reads the chat, gets in everywhereKeep them in a password manager
Card, bank, Social Security or passport numbersThe chatbot does not need them to answerReplace them with "XXXX"
Screenshots of documents or emailsThey often show addresses, account numbers and other people's names you did not noticeCrop or blur first, or type out only the part you need
Other people's health, legal or money detailsThey did not agree to share themDescribe the situation without names
Client files or unreleased business plansPersonal accounts are not covered by your company's contractsUse the business plan your company pays for, or nothing

Don't

  • Don't rate answers with thumbs up or down in a conversation that contains personal details: with ChatGPT and Claude, feedback can send the conversation into training or keep it for years.
  • Don't install browser extensions or apps that promise "free GPT" or "AI for all sites". Many read every page you visit.
  • Don't assume a shared chat link is private. Anyone with the link can read it.

Running a business? Add a line about AI tools to your staff rules: which ones are approved, and what never goes into them. It fits next to the other items in our small business cybersecurity checklist. And because AI also powers the newest scams, make sure your family knows the warning signs in our guide to AI voice clone scams.

How we checked this

Written by Ariel, ICT engineer. On 2026-10-05 we compared every setting name, menu path and retention period with the companies' own help pages: OpenAI's data controls, temporary chat and retention articles, Google's Gemini Apps Privacy Hub and activity help, Anthropic's privacy center and incognito chat article, and Microsoft's Copilot privacy controls. Menu names change often and can differ by country and app version, which is why each step names the setting rather than relying on screenshots. This page contains no affiliate links.

Sources