Temporary Chat

What is Temporary Chat?

Temporary Chat is a per-conversation mode in consumer AI assistants that prevents a single session from being saved to your chat history, from reading or writing your saved memory, and from being used to train the underlying model. It is the deliberate opposite of the persistent memory features that ChatGPT, Claude, and Gemini have rolled out since 2024. Where a normal conversation can both draw on what the assistant already knows about you and add new facts to that store, a temporary conversation starts from a clean slate and leaves no lasting trace in your account profile.

Each major vendor ships its own version of this idea under slightly different names. OpenAI calls it Temporary Chat in ChatGPT. Anthropic calls it an incognito chat in Claude. Google calls it a Temporary Chat in the Gemini app. The labels differ, but the core contract is consistent: the conversation does not appear in your recent or saved chats, it does not personalize future replies, and each vendor states it is excluded from model training.

The most common point of confusion is that off-the-record does not mean unrecorded. None of these modes is end-to-end encrypted, and the provider still receives the plaintext of your messages on its servers in order to generate a response. Each vendor retains temporary conversations for a stated window (up to roughly 30 days for ChatGPT and Claude, and up to 72 hours for Gemini as of mid-2026) for safety, abuse detection, and to process any feedback you submit. Temporary Chat is a memory and personalization control, not an anonymity or zero-retention guarantee.

• A per-conversation toggle that turns off history saving, memory read/write, and training contribution.
• Branded as Temporary Chat in ChatGPT and Gemini, and as an incognito chat in Claude.
• Does not appear in recent or saved chats and does not personalize later responses.
• Not end-to-end encrypted: the provider still receives and briefly retains the messages.
• A recall-off control for the session, not an anonymity or zero-retention feature.

How does Temporary Chat work across ChatGPT, Claude, and Gemini?

In ChatGPT, you start a Temporary Chat by clicking the model name at the top of the window and selecting the Temporary Chat option (a similar toggle exists in the desktop and mobile apps). OpenAI's help documentation states that Temporary Chats do not use your saved memories and do not create new ones, do not appear in your history, and are not used to train its models. OpenAI states these conversations are deleted from its systems within 30 days, subject to the legal-hold caveat described later on this page.

In Claude, the equivalent is an incognito chat, started by clicking the ghost icon in the upper right corner when you begin a new chat outside a project. Anthropic's support documentation states that incognito chats are not saved to your chat history, will not be pulled in when Claude searches previous conversations, do not use Claude's existing memory, and will not be included in future memory summaries. Anthropic states they are not used for training and are retained for 30 days by default, or longer where an Enterprise organization has set a custom data retention policy.

In the Gemini app, Temporary Chats began rolling out on August 13, 2025, and are reachable by opening the side drawer and tapping the outlined chat icon next to New chat. Google's blog states that Temporary Chats will not appear in your recent chats or Gemini Apps Activity, will not be used to personalize your Gemini experience, and will not be used to train Google's AI models. Google states these chats are kept for up to 72 hours so it can respond to you and process any feedback you choose to provide.

A shared trait across all three is that the mode is one-way for that session. Once you are in a temporary or incognito conversation you generally cannot convert it into a saved chat or reopen it after closing, because the whole point is that nothing persists in your account.

• ChatGPT: toggle Temporary Chat from the model-name menu; deleted within 30 days per OpenAI.
• Claude: click the ghost icon to start an incognito chat; retained 30 days by default.
• Gemini: open a Temporary Chat from the side drawer; kept up to 72 hours per Google.
• All three exclude the session from memory, history, personalization, and training.
• The session is one-way: you usually cannot save or reopen it once closed.

What does Temporary Chat NOT guarantee about privacy?

Temporary Chat narrows what the assistant remembers, but it is not a privacy shield in the strong sense. None of the consumer implementations is end-to-end encrypted, so the provider's servers process your messages in cleartext to generate each reply. The vendor can therefore read the content during the retention window, and that window exists precisely so the company can run safety and abuse-detection checks. If you would not want a message stored on a provider's infrastructure at all, Temporary Chat does not change that calculus.

Retention is the detail people most often get wrong. Not saved to history means the conversation is absent from your visible account, not that it is instantly erased everywhere. ChatGPT and Claude state a typical retention of around 30 days, and Gemini states up to 72 hours, after which the vendor says the data is deleted. Legal obligations can override these defaults. In the New York Times copyright litigation, a U.S. court ordered OpenAI to preserve consumer ChatGPT and API logs that it would otherwise have deleted. OpenAI has said that broad indefinite-retention obligation ended on September 26, 2025, but that it continues to securely store a limited set of historical data covering roughly April to September 2025 because of the ongoing case. In January 2026 the district court separately affirmed an order requiring OpenAI to produce a 20 million-conversation sample of those logs to the plaintiffs, which OpenAI has continued to contest.

Temporary Chat also does nothing about metadata or the surrounding environment. Your account, IP address, device, and the fact that a request was made can still be logged, and anything you paste into the box still leaves your device. The mode controls assistant memory and history, not network-level or account-level visibility. It is best understood as the recall-off flip side of the memory conversation rather than a substitute for a privacy-focused tool. Systems like MemX take a different posture, keeping a user's memory private by architecture through per-user isolation and encryption at rest, but that is a design choice about a stored memory layer, not the same thing as a transient off-the-record chat.

• Not end-to-end encrypted: the provider processes your messages in plaintext.
• Retained for a stated window (around 30 days for ChatGPT and Claude, up to 72 hours for Gemini) before deletion.
• Legal holds can override default deletion, as in the OpenAI versus New York Times litigation over preserved ChatGPT logs.
• Account, device, and IP metadata can still be logged outside the chat content.
• It is a memory and history control, not anonymity and not a zero-knowledge guarantee.

When should you use Temporary Chat (and when not)?

Temporary Chat is most useful when you want a clean, throwaway session that will not affect the assistant's long-term picture of you. Good cases include one-off research on a topic unrelated to your usual work, asking a sensitive personal question you do not want surfacing in later replies, drafting something for a context the assistant should not generalize from, or testing how the model behaves without your accumulated memory steering it. Because memory is off, the answers are also more neutral and less anchored to your saved preferences, which can help when you want an unbiased baseline.

It is the wrong tool when you actually want continuity. If you are building an ongoing project, want the assistant to recall your style, stack, or prior decisions, or expect to return to the thread later, a temporary session works against you because nothing is kept. The same applies to any workflow where you rely on the assistant referencing earlier conversations, since a temporary chat is walled off from both reading and contributing to that history.

Importantly, do not reach for Temporary Chat as a way to share secrets you would not entrust to the provider. Because the content is still received and briefly retained on vendor servers and is not end-to-end encrypted, highly sensitive credentials, regulated data, or anything you need cryptographically protected should not go into any consumer assistant, temporary mode included. Use it to manage memory and personalization, and use dedicated controls or different tools for true confidentiality.

• Use it for one-off, off-topic, or sensitive questions you do not want shaping future replies.
• Use it to get a neutral baseline answer free of your saved memory and preferences.
• Avoid it when you want continuity, project context, or recall of earlier threads.
• Do not treat it as a vault for secrets, credentials, or regulated data.
• Pair it with deleting memory or disabling personalization for broader, lasting control.

How is Temporary Chat different from turning memory off or deleting history?

Turning off memory is an account-wide setting, while Temporary Chat is a per-conversation mode. In ChatGPT, for example, you can disable saved memories entirely under Settings and Personalization, which stops the assistant from creating or referencing memories across all of your chats. Temporary Chat is narrower and more surgical: your normal chats keep using memory as usual, and only the single temporary session is excluded. People often choose Temporary Chat precisely because they want memory on most of the time and off just for one conversation.

Deleting history is a different action again. Removing past conversations or clearing saved memories cleans up what already exists, but a new normal chat will start logging and personalizing again right away. Temporary Chat is preventive rather than corrective: it stops anything from being saved or learned in the first place for that session, so there is nothing to delete afterward. The two approaches complement each other, and a thorough privacy routine often combines a memory-off setting, periodic history and memory deletion, and Temporary Chat for individual sensitive sessions.

Temporary Chat is also distinct from a vendor's broader data-control settings, such as opting out of having your conversations used to improve or train the model. Those opt-outs govern the training pipeline for your regular chats but still leave the conversations in your history and in memory. Temporary Chat bundles the no-training behavior together with no-history and no-memory for a single session, which is why it is the cleanest single switch for an isolated, low-footprint conversation, while the account-level controls remain the right tool for changing default behavior everywhere.

• Memory-off is account-wide; Temporary Chat is one conversation at a time.
• Deleting history is corrective cleanup; Temporary Chat is preventive and leaves nothing to delete.
• Training opt-outs cover regular chats but still keep history and memory; Temporary Chat removes all three for the session.
• Combine memory-off, periodic deletion, and Temporary Chat for layered control.
• Choose Temporary Chat when you want memory on by default but off for one thread.