Google Keeps Gemini Chats for 2 Months — Even With History Off

The first time I turned off Keep Activity in Gemini, I felt a small, almost childish sense of victory. For three days only, nothing I asked the chatbot would be kept. It was a clean, simple promise: I am gone in 72 hours.

Old key and laptop, security and data privacy theme
Image: Pudelek via Wikimedia Commons (CC BY-SA 4.0)

Then Google changed the rules, quietly, on November 17.

The Numbers

Google is extending how long it keeps temporary Gemini chats — the conversations you make when chat history is turned off — from 72 hours to around two months. It says the change is aimed at improving safety and security as AI systems become more capable, and it will roll out globally in the Gemini app over the coming weeks.

That is the news. The context is what I want to sit with for a minute.

Google’s updated Gemini Apps Privacy Hub now says future chats will be saved “for around two months,” used “to respond to you, process your feedback, and protect Google, its users, and the public.” The Register, covering the same update, put the scale in its headline: users who try to opt out of tracking will still see their data held 20x longer.

Let me pause on that. Three days became two months. Twenty times longer. I do not dispute the reason — a chatbot that remembers enough context to answer coherently needs some memory, and a longer temporary window does make the next answer slightly better. But around is doing a lot of work, and users deserve to know what that window actually covers.

What Actually Changed

The retention floor was only half the story. Along with it, Google is splitting the single Keep Activity toggle into two separate settings: one for how long your chat history is saved, and another for whether your conversations may be used to train generative AI models.

Before this, the two were bundled. Turn off Keep Activity and you stopped Google from training on your chats — but you also wiped your visible Gemini History. One switch, two decisions, and the privacy choice doubled as the archival choice.

Now you can keep your archive and still tell Google not to train on it. That is a genuine improvement in control, the kind of granular setting I have been asking for. It also means keeping your history and opting out of model training are finally separable. Google’s own blog frames it as more control over when your history is used for training, and that framing is fair.

What did not change is the bigger story. Even with history off, future chats are still stored for about two months and are still used to respond to you and to protect Google, its users, and the public. The temporary path is not used to train AI models, the company says — but reviewed chats and related data can be retained for up to three years, even if you delete your history later.

The Fine Print Nobody Reads

Here is the bit that should bother the privacy-conscious. The three-year figure is not exactly new — reviewed chats and related data have been kept for up to three years for a while. What is new is that the temporary, history-off window is now two months, and the gap between setting says off and Google still reads my messages has widened.

The same documentation notes that submitting a feedback request on a post can extend the temporary retention to three years. Three years. I have sat in rooms where that question — how long do we keep a log? — gets debated over coffee, with backup tapes, retention policies, and audit trails. Users get the abbreviated version of that conversation, and the answer is ten times the length of the default.

For the record, the visible Gemini History feature still defaults to 18 months, adjustable to three months, three years, or indefinite. So off now covers two different things: your visible archive (18 months, unless you change it) and your temporary retention (about two months, with a floor you cannot adjust). Neither is zero.

The Competitors and the Floor

I checked how Google’s window compares to the others, because longer is meaningless without a yardstick. OpenAI says it may keep temporary ChatGPT chats for up to 30 days. Anthropic retains incognito Claude chats for 30 days by default. Google’s two months sits about twice as long as those, a fact Gizmodo lands on in its coverage.

Twice as long is not a clean win. It means the average user, who mostly just wants a tool that does not archive them, is next to no closer to a clean answer than they were six months ago.

My Take, As Someone Who Builds and Manages Systems

I do not use Gemini as much as I used to. I turned history off years ago on most of the AI agents I trust with my code, and I’d rather use a model that does not store my questions in the first place. That preference is not just tech-bro minimalism; it is the same calculus my ICT division applies to every system we run. Who can see what, when, for how long, and what can be retrieved? The answer is a design document, not a setting.

As an ICT division manager, I watch data liability the way a boxer watches a jab. When you hand a chatbot your questions, you are also handing it context about your work, your plans, the things you are trying to figure out. That is valuable data, and it stops being yours the moment you type it. Google’s update does not change that, exactly — but it does make the off switch feel less like a wall and more like a floor. You are still in the building.

The burden of the fine print falls hardest on people who never thought they needed to read it. The average user does not have a data-privacy policy on their desk. They have a chat box. And that chat box now remembers them about 20 times longer than it did last month.

There is a wider pattern worth naming. This is not an isolated decision; it is part of a clear direction, and I mean that as an observation, not a smear. AI companies are doubling down on memory, not thinning it. Every feature that makes an assistant feel native — remembering your preferences, your tone, your past questions — is a feature that makes it remember you. Big Tech is here to let us down, and the best we can do is read the label. The off toggle is a concession to the fact that some people want privacy, not a guarantee that privacy will be honoured.

There is a reason the split controls matter. They are Google saying, in effect, we know you do not trust us, so we will let you separate the memory from the learning. That is real progress, and it is worth praising. But it is a patch on a larger trend, not a reversal of it.

What I Would Do

If you are asking what this means for you, here is the honest bottom line. First, do not mistake the new split for a reset. Second, treat off as a floor, not a wall: your chats are still being read, for about two months, and reviewed feedback can linger for three years. Third, if zero retention matters to you, do not use the tool — there are models and settings that keep nothing on their servers. Fourth, for anyone managing AI tools in a team or institution, this is a reminder — vet any AI tool you trust with your code — that your data-handling defaults are whatever the vendor sets as the default, and defaults are public, not secret.

I am not saying Google is lying. To respond to you, process your feedback, and protect Google, its users, and the public is a mouthful, but it is not a lie. It is a safety and reliability decision dressed up as a privacy promise, and the two-month window is the honest middle ground: long enough to work, short enough to laugh about later.

Here is what I want you to take away, so you do not have to be a privacy engineer to walk away with the right assumption. If you turn history off, assume your chats are still being read, for about two months. If you want zero retention, do not use the tool. And if you build tools people use, treat the user turned it off as a starting point for trust, not a finished guarantee.

The control is real. The data is not gone. The promise is around, and around is the word that should make you pause.

Filed under Tech & Gadgets
Last Update: November 18, 2026 by Felix AlterEgo
0 0 votes
Article Rating
Subscribe
Notify of
guest

This site uses Akismet to reduce spam. Learn how your comment data is processed.

0 Comments
Newest
Oldest Most Voted