Do ChatGPT, Claude and Gemini train on my data?

Updated October 17, 2026 · from live checks of 44,798 MCP servers · Methodology

On consumer plans, ChatGPT, Gemini, Microsoft Copilot and Perplexity can use your chats for training unless you opt out, and Claude does only if you leave its model-improvement setting on; business and API plans from OpenAI and Anthropic don't train on your data by default.Verified against vendor help pages on 2026-10-17.

Training on your chats, by product

Product / planTrains on your chats?How to stop itSource
ChatGPT (Free, Plus, Pro)Yes, unless you opt outSettings → Data controls → “Improve the model for everyone” off, or Temporary ChatOpenAI
ChatGPT Business / Enterprise / Edu, OpenAI APINo, by default—OpenAI
Claude (Free, Pro, Max)Only if “Help improve Claude” is onSettings → Privacy → model improvement off, or Incognito chatsAnthropic
Claude Team / Enterprise, Anthropic APINo, by default—Anthropic
Gemini app (personal Google account)Yes, unless you opt outTurn off “Keep Activity”, or use temporary chatsGoogle
Microsoft Copilot (consumer)Yes, unless you opt outProfile → Privacy → “Training on conversation activity” offMicrosoft
Perplexity (consumer)Yes, unless you opt outSettings → “AI Data Retention” / AI data usage offPerplexity

What about connected apps (MCP)?

When you connect an app through MCP, the data a tool returns becomes part of your chat, so the same training setting applies to it. Turn training off, or use a business plan, before connecting work accounts.

Do temporary or incognito chats get used for training?

No. ChatGPT Temporary Chats, Claude Incognito chats and Gemini temporary chats aren't used for training, per each vendor.

Does opting out delete past chats?

No. Opting out applies to future chats. Delete old chats separately.

Try “Notion”, “Stripe”, “create invoice”, “browser”