Claude Privacy Settings: A Real Walkthrough
5 min read
Every assistant asks you to trust it with something. The honest way to decide how much is to know the specific controls, which ones are on by default, and where the edges are. This article walks through the privacy settings Anthropic documents for consumer Claude plans (Free, Pro and Max), plus memory and incognito, in the order you would meet them.
One boundary first. Anthropic's privacy page covers consumer plans and Claude Code used from those accounts. Commercial products, meaning Team, Enterprise and the API, are covered by separate documents with different terms, so if you use Claude through an employer, read your organization's policy and ask your administrator before you assume anything here applies. For the basics of using Claude, the Complete Beginner's Guide to Claude is the place to start.
The question people actually ask: is my chat used for training?
Anthropic's page says your chats and coding sessions can be used to improve its models in three situations:
- You have turned on the model improvement setting.
- The conversation was flagged for safety review.
- You opted in some other way, for example by joining the Trusted Tester Program.
When a conversation is used, it is the full related conversation, plus any custom styles or preferences, and data collected through Claude for Chrome.
Look at the first item. It means there is a setting you control, so find it and look at it. Open your privacy settings and check the state of the model improvement option today rather than relying on a memory of what you chose long ago.
Flagged conversations are separate
Even with the setting off, a conversation flagged for safety review may be used or analyzed to improve detection and enforcement of the usage policy, including training models used by Anthropic's safeguards team. That is a different path from the setting, and the setting does not switch it off.
What is not included
Two exclusions in the same article are useful to know.
- Incognito chats are not used to improve Claude, even if model improvement is enabled.
- Raw content from connectors, including remote and local MCP servers such as Google Drive files, is excluded. But the page warns that such content can still be included if you copy it directly into a chat. The route matters: letting Claude read a file through a connector is treated differently from pasting that file into the message box.
Feedback is a separate door
When you click thumbs up or thumbs down, Anthropic stores the full related conversation in a secured back end for up to five years. It says the feedback is de-linked from your user ID before use, may be used to analyze service effectiveness, conduct research, study user behavior and train AI models where permitted by law, and is not combined with your other conversations.
Practical reading: a thumbs-down on a response is a decision to share that whole conversation. If the chat contained anything private, think before you click.
Memory and chat history, as privacy controls
Memory has its own controls, covered in What Claude Remembers Between Conversations. The ones that matter for privacy:
- Memory does not save sensitive topics such as health, religion, politics or gender identity by default, and some information is never saved, including government ID numbers, criminal history, financial account numbers and immigration status.
- You can pause memory, reset it permanently, turn it off for one chat before you send the first message, or review and delete individual memories.
- Deleting a conversation does not delete the memories created from it. Delete those separately.
- Incognito chats, started with the ghost icon in a new chat outside a project, are not saved to memory or to your chat history, and are available on every plan.
Web search and location
If you use web search, Anthropic notes that Claude may use your location, inferred from your IP address, to localize results. That is a convenience for local questions, and a fact worth knowing if your location is sensitive. When Claude Is Searching the Web and When It Isn't covers the search side.
A twenty-minute review, in order
- 1
Check the training setting
Open your privacy settings and read the model improvement option. Decide whether it should be on, and set it deliberately.
- 2
Review memory
Go to Settings, then "Your Claude," then "Manage" next to "Your memory." Read every topic. Delete anything you would not want repeated in a draft.
- 3
Look at your connectors
Under Customize, then Connectors, check what is connected and whether each one can only read or can also change things. Disconnect what you do not use.
- 4
Choose a default for sensitive work
Decide which kinds of conversations should be incognito from the start, and make that a rule: legal questions, health questions, anything about a person who is not you.
- 5
Review your habits, not just your settings
Settings cannot protect what you paste. Look at the last ten things you pasted into Claude and ask whether you would have emailed them to a stranger.
What to keep out of a chat in the first place
Settings reduce risk but they do not remove it. A simple rule is stricter than any setting: before you paste, ask whether you could remove the identifying details and still get the help you need. Replace names with roles ("the client," "the employee"), figures with ranges, and account numbers with placeholders. You lose nothing in the answer and you hold back what mattered.
Work data is a different category
If you use Claude for your job, your employer's rules apply on top of Anthropic's. Using a personal account for client files, patient information or confidential documents can break those rules even when the privacy settings are perfect.
If you use more than one assistant, each has its own controls: Copilot Privacy Settings: A Real Walkthrough and Grok Privacy Settings: A Real Walkthrough cover those.
Official sources
- Is my data used for model training? | Claude Privacy Center
- Use Claude's chat search and memory to build on previous context | Claude Help Center
- Enabling and using web search | Claude Help Center
Checked on October 24, 2026. Features, plans and names change often, so the vendor's own pages are the final word.