Is my data safe?

Last checked 2026-10-15

Most consumer AI chatbots may use your conversations to improve their models unless you turn that off. Business and enterprise plans usually don't train on your data by default. Whatever the setting, avoid pasting passwords, ID numbers, health records or other people's private data unless your organisation has approved the tool.

Quick checklist

Find the training/“improve the model” setting and decide whether to turn it off.
Use temporary, incognito or private chats for sensitive questions where offered.
Be aware that thumbs-up/down feedback can send that conversation for training even if you opted out (ChatGPT, Gemini and Grok docs).
Delete old chats you don't need.
For work, use your employer's approved plan.

ChatGPT (incl. Work & Codex)

Setting: Settings → Data controls → turn off “Improve the model for everyone”.

When it's off, new conversations aren't used to train OpenAI models; they still stay in your chat history (turning it off deletes nothing). Signed-in: the setting follows your account across devices; signed-out: it's saved in that browser. Temporary Chats aren't used to improve models but may be kept up to 30 days for safety. Thumbs-up/down feedback can let that whole conversation be used for training. Business, Enterprise, Edu and ChatGPT for Healthcare content isn't used for training by default.

Sources: help.openai.com/en/articles/7730893-data-controls-in-chatgpt · checked 2026-10-15 (official help page, fetched and checked 2026-10-15)

Claude (incl. Claude Code)

Setting: Settings → Privacy → turn off “Help improve our AI models” (claude.ai/settings/data-privacy-controls).

Covers Free, Pro and Max, including Claude Code on those accounts. When off, new chats and coding sessions aren't used for future training, but data already in training runs or trained models stays there. Chats flagged by safety classifiers may still be used for trust-and-safety work. Retention: 30 days if off, 5 years if on (stated in Anthropic's Claude Code data-usage docs for consumer accounts). Commercial plans (Team, Enterprise, API) aren't trained on unless the customer opts in.

Sources: privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings · code.claude.com/docs/en/data-usage · checked 2026-10-15 (official help page, fetched and checked 2026-10-15)

Google Gemini

Setting: Open Gemini Apps Activity (in gemini.google.com: Settings & help → Activity) and turn off “Keep Activity”.

While Keep Activity is on, Google uses your chats to improve its services, including training AI models, and some are reviewed by people; reviewed chats are kept up to 3 years even if you delete your activity. Activity auto-deletes after 18 months by default (you can pick 3 or 36 months, or never). With it off, future chats aren't used for training unless you send feedback, but are kept with your account for 72 hours. Temporary chats aren't used to train Google's AI models. Some Connected Apps (e.g. Workspace) don't work with it off.

Sources: support.google.com/gemini/answer/13278892?hl=en · support.google.com/gemini/answer/13594961 · checked 2026-10-15 (official help page, fetched and checked 2026-10-15)

Microsoft Copilot

Setting: Older Copilot app: Profile → Privacy (web) / Settings → Privacy (Windows/macOS) / Account → Privacy (mobile) → turn off “Training on conversation activity” and “Training on voice conversations”. New Copilot app (from 18 Aug 2026): Microsoft's privacy-controls page lists Memory, shared experiences, chat history, web search and ad settings under Settings → Personalization, but doesn't describe a model-training toggle; check that page for updates.

For the older app, Microsoft says opting out excludes future conversations from model training, but not from other product improvement, advertising, safety or compliance uses; conversations are kept 18 months by default. It says it doesn't train on work Entra accounts, signed-out users, signed-in under-18s, Microsoft 365 Personal/Family app users, or users in Brazil, China (excl. Hong Kong), Israel, Nigeria, South Korea and Vietnam. Some human review happens and can't be opted out of. We could not confirm the training policy for the new app.

Sources: support.microsoft.com/en-us/privacy/microsoft-copilot/privacy-controls · support.microsoft.com/en-us/microsoft-copilot/microsoft-copilot-privacy-controls · support.microsoft.com/en-us/microsoft-copilot/privacy-faq-for-microsoft-copilot · checked 2026-10-15 (official help page, fetched and checked 2026-10-15)

Grok

Setting: Grok app: Settings → Data Controls → “Improve the model”. grok.com: Settings → Data → “Improve the Model”.

When off, new conversations aren't used for training; feedback you send may still be. Private Chat isn't used for training and is deleted within 30 days; deleted conversations are removed within 30 days (with exceptions for de-identified data or safety/legal reasons). Business and enterprise content isn't used. Grok inside X follows X's own settings. In some regions (not EU/UK), logged-out use has no opt-out.

Sources: x.ai/legal/faq · checked 2026-10-15 (official help page, fetched and checked 2026-10-15)

Perplexity (incl. Comet)

Setting: Account settings → Preferences → turn off “AI data retention”.

“AI data retention” is on by default for Free, Pro and Max. Turning it off applies only to data collected afterwards; previously collected training data can't be removed. Data may still be processed for running the service, legal compliance and product improvement. Enterprise data is never used for training.

Sources: www.perplexity.ai/help-center/en/articles/11564572-data-collection-at-perplexity · checked 2026-10-15 (official help page, fetched and checked 2026-10-15)

Meta AI

Setting: Meta AI app → Menu → Settings → Data & privacy → Manage your information → Delete all chats and media.

This deletes chats and media you created (public posts aren't deleted). Meta's help page says it uses your interactions with AIs to improve AI at Meta, and may share messages with select partners when Meta AI can't answer. We found no general training opt-out on that page. EU/UK objection options weren't verified.

Sources: www.meta.com/help/artificial-intelligence/2457110494637611/ · checked 2026-10-15 (official help page, fetched and checked 2026-10-15)

Other tools on this site (Mistral, DeepSeek, Cursor, GitHub Copilot, n8n and others) haven't been checked for this guide yet; see each vendor's privacy policy.