// how to
How to Keep Your Data Private in Claude: Model Improvement and Incognito Chats
In Claude’s Privacy Settings, check the Model Improvement setting: Anthropic uses consumer chats for training only if it’s on (or a chat is flagged for safety review). For sensitive one-off questions, use an Incognito chat (the ghost icon in a new chat): Anthropic says those aren’t used to improve Claude, even with Model Improvement on.
The short version
Check one setting, use one feature, and keep one habit. The setting: Model Improvement. The feature: Incognito chats. The habit: never paste what you wouldn’t email to a stranger.
Step by step
1. Check Model Improvement
Open Claude’s Settings > Privacy (Anthropic links it as claude.ai/settings/data-privacy-controls) and look at Model Improvement. Anthropic says consumer chats and coding sessions are used for training if you opt in there, if a conversation is flagged for safety review, or if you join a program like Trusted Tester. Turn it off if you don’t want to contribute.
2. Use Incognito for sensitive questions
Start a new chat (not inside a Project) and press the ghost icon. Per Anthropic’s Help Center, Incognito chats aren’t saved to your history or Claude’s memory, and Anthropic says they’re not used to improve Claude even if Model Improvement is on. They’re still kept for a limited time for safety.
3. Manage memory
Memory helps Claude remember useful context across chats. If you’d rather it didn’t, check memory in Settings, or switch it off for a single chat from the + menu before your first message.
4. Think before rating
Thumbs up or down feedback is stored for up to 5 years, according to Anthropic. Don’t rate chats that contain sensitive details.
Never paste
- Passwords, recovery codes or API keys
- Full card or bank numbers, Social Security or ID numbers
- Other people’s medical, legal or financial details
- Confidential company documents you’re not allowed to share
Prompts to copy
Work with placeholders:
I've replaced names and numbers with [CLIENT], [AMOUNT] and [DATE]. Keep the placeholders in your answer.
[paste text]
Mistakes to avoid
- Assuming Incognito works inside Projects. It’s for standalone chats only.
- Rating sensitive chats. Feedback is kept much longer.
- Using a personal account for company data when your employer has rules or a Team plan.
Good to know
Team and Enterprise plans have their own terms; for example, Anthropic notes Incognito chats there can be included in organizational data exports. Settings and names can change; search Anthropic’s Privacy Center if a menu has moved.
Related: use Claude for your small business.
Frequently asked questions
Does Anthropic train on my Claude chats?
According to Anthropic's privacy article (March 16, 2026), consumer chats are used for training when you opt in through the Model Improvement setting, when a conversation is flagged for safety review, or when you join a program like Trusted Tester. Incognito chats aren't used to improve Claude.
What's an Incognito chat?
A chat that isn't saved to your history or Claude's memory. You start one from a new chat (not inside a Project) with the ghost icon. Anthropic says it still keeps such chats for a limited time for safety.
What about thumbs up and down feedback?
Anthropic says feedback you send with those buttons is stored for up to 5 years, so don't rate a chat that contains something sensitive.
Free download
Get the AI Coding Starter Kit
A PDF you can use with any AI coding assistant: a project-rules template, 25 tested prompt patterns, a team adoption checklist and an assistant comparison sheet. The editable Markdown sources are included.
- Works with Copilot, Claude Code, Codex, Cursor and more
- Double opt-in, one-click unsubscribe
- No trackers, never shared