ChatGPT Lets Temporary Chats Use Memory, Then Convert Them to Regular Chats

ChatGPT users can now bring their existing setup into a conversation that remains disposable—until they choose to save it. That new save action converts a Temporary Chat into a regular chat, changing its place in history and the settings that govern memory and model improvement.

OpenAI updated Temporary Chat on August 27. Users can decide before their first message whether the session may use existing memories, custom instructions, and plugins. The choice is set at the start of the chat; it cannot be changed once the conversation has begun.

Context without a new record

The change breaks apart choices that were formerly tied together: whether ChatGPT may draw on prior personal context, whether the current exchange writes new context, and whether the exchange stays in account history. Previously, a user who found a Temporary Chat worth keeping had to abandon it or manually move its contents into a standard conversation.

The three controls now operate separately

• Personalization: A user may allow a Temporary Chat to use existing memories, custom instructions, and plugins before sending the first prompt.
• Memory creation: A personalized Temporary Chat can read existing memories, but it cannot create or update them while it remains temporary.
• Persistence: Leaving the chat unsaved keeps it outside history; saving the full exchange turns it into a regular chat.

Save changes the governing rules

An unsaved Temporary Chat stays out of chat history, does not create memories, and is not used to improve OpenAI’s models. Those protections apply whether the user chose a blank-slate session or permitted it to read existing context.

Temporary does not mean immediately deleted

Temporary status has a retention limit rather than an instant-erasure promise. OpenAI may retain a copy of an unsaved Temporary Chat for up to 30 days for safety purposes. That period is distinct from the chat’s absence from history, memory creation, and model-improvement use.

Saving ends the temporary treatment for the complete exchange. It appears in history and may become available to ChatGPT’s personalization systems under the user’s normal settings. A user who has disabled model training keeps that preference after saving, so conversion does not override account-level controls.

Third-party actions sit outside the simple boundary

The new flexibility also makes the moment of choice more consequential. A user can bring familiar preferences into a short-lived session without adding new memories, then preserve useful work later. But information sent through third-party actions may be governed by the recipient’s privacy policy and may be retained beyond OpenAI’s 30-day safety window.