Google announced a redesigned set of Gemini privacy controls on November 16. The update breaks apart choices that had shared one switch: whether you can return to old conversations, whether future chats may train Google's generative AI models, and how long Google may keep chat data for other purposes. Those choices sound similar on a settings page. They have different consequences. A person who turns off model training may still have chats stored in visible history. A person who turns off history will no longer see future conversations there, but Google says it will keep them with the account for around two months.

The new controls are rolling out globally over several weeks, so some accounts may still show the old interface. They apply to personal Gemini accounts. Work and school accounts may operate under terms and retention choices controlled by a Workspace administrator.

One switch used to do two jobs

Gemini's old "Keep Activity" setting joined visible chat history to permission for model training. To keep future conversations out of generative-model training, a user also had to give up the ability to return to them later. Ars Technica reported that the update removes that tradeoff.

Google's November 16 announcement says, "We're giving you more control over when your Gemini chat history is used to train generative AI models. New settings let you keep your Gemini chat history, while opting out of generative AI model training." That is a useful improvement. Someone can preserve a research trail or ongoing project without offering future chats for model training. The same announcement, however, also changes the storage period for conversations made with history off and for temporary chats. Understanding that change requires reading the new controls as three separate questions.

History controls what comes back to you

For personal accounts belonging to people 18 and older, Gemini History is on by default. Google's help page says saved history is set to delete automatically after 18 months. A user can shorten that period to 3 months, extend it to 36 months, or turn automatic deletion off. When History is off, future conversations do not appear in the user's Gemini History panel. That describes what the account holder can return to later. It does not set an immediate deletion deadline for Google's systems.

The training choice covers future model use

A separate setting governs whether Google uses future chat content to train generative AI models. Google's current help page says this training control is off by default and can be enabled only while History is on. The content covered can include text, uploads, Gemini's responses, tasks, tool calls, location information, and data from connected apps.

The choice works prospectively. Turning training off does not pull back material already reviewed or used under previous settings. Deleting a conversation from visible history also has a different function from changing the training choice.

There is a further exception worth knowing. Google's Gemini Apps Privacy Hub says that if History is off and a user submits feedback, Google may collect the feedback, the last 24 hours of related chats, and included attachments or connected-app content to understand it.

Retention continues after history is off

Google says conversations made with History off are saved with the account for around two months. The same period applies to temporary chats. Google says it keeps them to respond to the user, process feedback, handle possible system failures, and protect the company, its users, and the public.

That period used to be much shorter. TechCrunch compared the current policy with an archived November 12 version and found that the stated period had increased from 72 hours to around two months. Google pointed the publication to its announcement but did not answer questions about what prompted the change or whether access and review would change.

Temporary chats still serve a clear purpose. Google states that they are not used to train its AI models. They also do not appear in recent chats or influence future personalization. Their name does not mean that the underlying data disappears when the window closes.

Human review has another clock. Google says a subset of chats is reviewed by trained reviewers, including service providers. It says chats are disconnected from the account before being sent to those providers. Reviewed chats and related data may be kept for up to three years, even when the user deletes history or later turns a setting off.

What the published policy cannot tell us

These sources establish Google's stated rules and the visible choices offered to users. They do not show how often an ordinary conversation is reviewed, what usually triggers access, or how often Google uses the longer retention window in a security investigation. The public material also does not independently test Google's process for disconnecting chats from user accounts before service-provider review.

The evidence supports a narrower conclusion. Gemini now gives users a better way to separate convenience from model training, while its history-off retention period has grown from roughly three days to roughly two months. Both facts can be true at once.

Check the setting that matches your concern

When the update reaches your account, open Gemini's Settings and then History. Check whether History is on, whether chat content and uploads may be used for training, and whether audio or Gemini Live recordings have a separate training setting. Review the auto-delete period, Connected Apps, saved public links, and any conversations you no longer need.

The policy itself offers the strongest rule for sensitive material: "Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services, including machine-learning technologies."

Before pasting a medical question, workplace document, financial draft, family photo, or another person's information into Gemini, ask which outcome matters to you: keeping it out of your history, keeping it out of model training, or minimizing how long any provider holds it. If the answer is all three, redaction or choosing not to enter it is safer than relying on a single switch.