
The Library Media — Science, Technology & Health · AI News & Intelligence Desk
‘History Off’ Now Means Two Months: Google Keeps Your Gemini Chats Longer in the Same Week It Asks to Run Your Errands
By Gideon Black
Google is extending the period it retains temporary Gemini chats and conversations made with chat history turned off from 72 hours to around two months, changing what users can expect when they choose not to keep a conversation in their visible history. The change, which Google says is rolling out globally in the Gemini app over the coming weeks, was reflected in updates to its Gemini Apps Privacy Hub on November 16 and 17.
The timing deserves attention. Google is expanding the capabilities of Gemini while also changing the privacy expectations attached to its least persistent chat option. The two developments do not establish a connection between the product’s new capabilities and the retention decision. But together they make the practical meaning of a user’s privacy settings an increasingly important question.
The change is substantial. TechCrunch, comparing the current Privacy Hub with its archived November 12 version, reported that the previous retention period was 72 hours. Google’s updated wording says temporary chats and conversations made with history off are stored with the user’s account for around two months. The Hub says this is so the chats remain available to respond to users, process feedback, protect Google, its users and the public, and address potential system failures.
Google’s explanation is not difficult to understand in principle. A service provider may need records to investigate abuse, respond to feedback or recover from technical problems. Google also says human reviewers assist in the relevant processes. Those are operational reasons for retaining information, not evidence that the company is using temporary conversations to train its models or that the change gives new people access to them.
That distinction matters because the updated policy does not say temporary chats are being used for model training. Google’s Hub says they are not. It also says that when history is off, future conversations are not used to train generative AI models if the user does not submit feedback.
The issue is therefore not a newly disclosed training use. It is the length of time a conversation can remain stored despite the user having chosen not to keep it in chat history.
At 72 hours, the setting offered a relatively short retention window. Around two months is a different proposition. A conversation can remain outside the user’s visible history while still being held by Google for weeks. That is not the same as an ordinary saved chat, but neither is it close to immediate deletion.
Google’s updated wording also points to a further retention distinction. Gemini Apps activity has an auto-delete default of 18 months, which users can change to three months, 36 months or indefinite retention. Separately, the Hub says conversations reviewed by human reviewers, together with related data, can be retained for up to three years. That material is disconnected from the user’s account and is not deleted when the user deletes their activity.
These are separate retention arrangements, and they should not be collapsed into a claim that every temporary conversation is kept for three years. The three-year period applies to chats that have been reviewed by human reviewers and related data, according to Google’s policy. But it does mean that deleting activity is not necessarily a guarantee that every related record has been erased.
The distinction between history and training is also changing. Previously, Google’s single “Keep Activity” setting linked whether Gemini conversations were retained in activity with whether they could be used to improve AI models. TechCrunch reported that Google has now separated the choices, allowing users to retain conversations in history without agreeing to their use for training.
For users who had previously disabled Keep Activity, both new settings are off by default. For users who had it enabled, chat history remains on while the training choice becomes separate. This is a meaningful improvement in control: users no longer have to choose between keeping a record of conversations and permitting their use for model training.
But the improvement does not reverse the retention extension. The settings answer different questions. One determines whether conversations remain in a user’s history; another concerns whether they can be used for training; the temporary-chat policy determines how long Google may retain conversations that are not kept in history. Treating those as interchangeable would leave users with an incomplete understanding of what their choices do.
TechCrunch reported that it asked Google what prompted the retention change and whether it would affect how conversations are accessed or reviewed. Google did not answer those questions, instead pointing the publication to its blog post.
That leaves important matters unresolved. The published information does not establish that the change expands who can review retained conversations. It does not establish a change to law-enforcement access or legal-process requirements. Nor does it establish a new training use for temporary chats. Google’s stated policy says the opposite on training, subject to the feedback condition it describes.
Those limits should be respected. A longer retention period is not, by itself, evidence of broader access or a new use of the data. But the absence of an answer about whether access or review practices have changed leaves users without clarification on a question that naturally follows from the extension.
The comparison with competing services provides some context, although the policies are not identical. TechCrunch reported that OpenAI says it may retain temporary ChatGPT chats for up to 30 days, while Anthropic retains Claude incognito chats for 30 days by default. The companies may retain some data for longer for legal or safety reasons, or when investigating policy violations. Google’s new period of around two months is longer than those stated default windows, but comparisons need to account for the different conditions and exceptions in each company’s policy.
The central issue is what users understand when they turn history off. The phrase suggests that a conversation will not become part of the continuing record associated with their use of the service. Google’s policy does not say that the conversation disappears immediately; it describes a period of storage for operational and safety purposes. That distinction is now more consequential because the period has grown from three days to around two months.
A setting can be accurately labelled and still be misunderstood if users do not distinguish between visibility, retention and training. Google’s separation of history from training makes those choices clearer in one respect. The longer temporary-chat retention makes the meaning of another choice less ephemeral.
The burden is not simply on users to read policy pages. When a company changes the practical consequences of a privacy setting, the explanation should make clear what is retained, for how long, for what purposes and under what review arrangements. Google has explained several reasons for the longer window. Its response to questions about what prompted the change and whether access or review practices are affected remains absent.
For now, the facts support a measured conclusion. Google’s temporary chats are not being described as training material, and the company has not disclosed a new access regime. But a user who turns history off should understand that the conversation may still be stored for around two months, and that a separate, longer retention rule applies to material reviewed by human reviewers. “History off” is not the same as “not retained”.
Gideon Black
Chief Skeptic, The Library Media
From The Library — the estate of our family.
Follow Us
