Learn · Your data · 8 min
What Meta AI, Muse, Grok, Copilot, ChatGPT, Gemini and Claude keep
Where each popular AI app keeps your chats, who can read them, and the switch that stops it training on you. Checked on 30 October 2026; menus move, so look again every few months.
Ask any AI app three questions
Does it learn from my chats unless I say no? Can a person at the company read them? How long does it keep them? And one more: if I press Share, who can see it? The answers below come from each company's own help pages, with the news stories linked at the bottom.
ChatGPT and Claude
On Free, Plus and Pro, your chats help train OpenAI's models unless you switch it off: Settings, Data controls, turn off "Improve the model for everyone". Business, Enterprise and Edu accounts are off by default. A Temporary chat is not used for training and is kept for up to 30 days, and deleted chats are removed within 30 days. A thumbs up or down can still send that whole chat for training even with the switch off, and memories and uploaded files are handled separately. Free and Go plans now show ads. Anthropic asked every Claude user to choose, by 8 October 2025, whether chats help train its models. Check your choice under Settings, Privacy, "Help improve our AI models". If you allow it, de-identified chats can be kept for up to five years; deleted chats go within 30 days. Memory is on by default. In July 2026 some shared Claude chats turned up in Google search, so treat any share link as public.
Gemini and Microsoft Copilot
While "Keep Activity" is on, Google uses your chats to improve its AI and some are read by human reviewers, including outside contractors. Reviewed chats are kept for up to three years, even if you delete your activity. To stop it: Settings and help, Activity, turn off Keep Activity; chats are then kept for 72 hours. Google's own page says not to type anything confidential. The free Copilot app learns from conversations unless you turn off "Training on conversation activity" and "Training on voice conversations" under your profile, then Privacy. History is kept for 18 months. Copilot inside Microsoft 365 apps for home, and Copilot on a work account, are not used for training. In September 2026 Microsoft folded its consumer Copilot into one work-focused app, so these menus may move.
Grok, Meta AI, Muse and Dot
Grok learns from your chats unless you turn off "Improve the model" in Settings, Data Controls, or use Private Chat (the ghost icon), which is deleted within 30 days. xAI says a limited number of its staff may read conversations. In 2025 more than 370,000 shared Grok chats were found in Google search. In 2026 Canada's privacy commissioner found Grok's image tool broke Canadian privacy law after it was used to make sexualised fake images of women and girls. Since 16 December 2025 Meta uses your chats with Meta AI to choose the posts and ads you see. In 2025 people published private chats to the Meta AI app's public feed without realising. Muse, Meta's agent app launched in September 2026, learns from you by default: turn off "Help improve our AI models" in Settings, Data controls. There is no switch to turn its memory off, though you can ask it to wipe what it holds. In October 2026 Meta paused Muse sharing after about 6,800 shared chats were found exposed. Dot, the companion app from New Computer, announced it was closing, with service until 5 October 2025. Its policy said it did not train on your chats, and it deletes your data if you ask. If you used it, send that request. Do not confuse it with "dots", the new agents OpenAI launched in September 2026.
The habit that beats every setting
Settings change; habits last. Keep health details, passwords, bank numbers, your children's names and schools, and other people's private information out of any chat. Use a temporary or private chat for anything sensitive. Never press Share on a chat you would not post publicly. Check these switches again every few months.
Try it now
Next, see the rights you have over your data, and how to ask a company what it holds about you.
Reviewed from: Data controls in ChatGPT (OpenAI), Temporary chat (OpenAI), Gemini Apps Privacy Hub (Google), Microsoft Copilot privacy controls (Microsoft), Is my data used for model training? (Anthropic), How long do you store my data? (Anthropic), Consumer FAQs (xAI), Improving your recommendations with AI at Meta (Meta), Muse, Meta's new personal AI agent (WIRED), Meta suspends Muse sharing (ProPakistani, citing Ars Technica), The Meta AI app is a privacy disaster (TechCrunch), Shared Claude chats may have ended up on Google (TechCrunch), 370k Grok chats made public (9to5Mac), Grok image tool and the privacy commissioner (CBC News). Written in our own words.
Drafted with AI from the linked sources. Check the links for the full detail.