Does “do not save this” actually prevent retention? Where should I verify settings and policies?

Full explanation

Check four things separately: chat history, memory for later use, training use, and retention by the provider/external services. Different switches may apply.

Claude says incognito chats do not enter history or memory and are not used for training, but default retention is still 30 days and organizational custom settings may be longer. OpenAI's API documentation says API data is not used for training by default, while abuse-monitoring logs generally may remain up to 30 days and some features retain application state; approved zero-data-retention controls have separate conditions. These concern a specific chat feature and an API service and cannot be interchanged. Claude incognito documentation, OpenAI API data controls

Verify in this order:

1. Identify the product, personal/organizational account and any third-party application.

2. Inspect privacy, training, history and memory settings and the actual mode of this conversation.

3. Read the feature's official policy for retention, deletion and exceptions; check administrator settings for organizations.

4. For connectors/external tools, check the recipient separately. OpenAI also notes that data sent to external MCP services follows third-party retention policies.

You can request:

Do not use this information for later personalization or write it into lasting memory or another file. Explain which product settings I should check. Before sending content to an external service, state the recipient and necessary scope.

This constrains task behavior without replacing settings checks. Remove irrelevant names/accounts when real identity is unnecessary. For submitted material, use actual deletion controls and check chats, files and memories separately; “I forgot” is not a deletion receipt.

References

1. Anthropic, Use incognito chats, official help

2. OpenAI, Data controls in the OpenAI platform, official API documentation