Gemini lets you turn off AI training without losing chat history, but storage rises to 2 months
Previously, to stop Google from training Gemini on your chats, you had to turn off history, and the old conversation could no longer be opened. Now the settings include two separate toggles: one for history and one for AI training. But with history turned off, chats are kept for two months instead of 72 hours. So it's better to protect your work data right away: turn off training manually in the Activity section.
AI-processed from Ars Technica; edited by Hamidun News
Imagine you pasted a piece of internal code or a contract draft into Gemini and want Google not to use it for training its model. Until now, that meant turning off chat history entirely, which meant yesterday's conversation could no longer be opened and continued. Now Google splits these settings into two independent toggles: one controls history, the other controls AI training on your conversations. But there is a downside, the one that made it into the headline: if you turn history off, Google will keep your chats not for 72 hours, as it does now, but for two months. Ars Technica reports this.
What changed in the settings
Previously the choice was rigid. To stop dialogues and attachments from being used to train Gemini, you had to turn off history entirely. It's like not wanting others to read your notebook, so you throw it away after every entry.
Now the two toggles work separately. Chat history still supports a configurable retention period, and model training can be turned off without losing access to earlier conversations. You can find the right switch in Gemini's settings under the Activity section. There is also a second route through your Google account: Data & privacy → My Activity → Other activity → Gemini Apps Activity. Google says the update is rolling out worldwide, but it may take several weeks to reach all users.
What falls outside the summary
The main caveat: training works on an opt-out basis. Google does not ask permission in advance, and until you find the toggle yourself, your chats may end up in training datasets. Opting out applies only to future conversations. Everything you have already sent to Gemini remains in the training data: prompts, model responses, uploaded files, tool calls, location, and data from connected apps. Chats that have already been reviewed and included in a training set are not removed retroactively.
Opting out of training does not mean Google can't see your data. The company may use it in other ways for up to three years; it simply won't add it to training sets. Google also says it tries to scrub personal information from history, but it offers no guarantees. You can delete any records manually on the same page, whatever settings you choose.
Why storage will grow
Today, with history turned off, Google still stores data for 72 hours, while the user can access it for 24 hours. After the update, the period will become two months. So access to the data will last longer, both for you and for Google. The company explains this as necessary to respond to you, process feedback, and protect Google, users, and society. In practice, this means Google will be better able to detect AI misuse even if a person has turned history off. Ars Technica calls the change a privacy plus, but with the caveat "mostly."
What to do in practice
If you or your employees send internal code, contracts, client data, or reports to Gemini, go to the settings and turn off training on the account. This no longer costs you your history, so there is less reason to delay. For corporate scenarios, it is more reliable to look at dedicated business plans and data-processing terms rather than the consumer app.
Product teams should note the approach itself: "memory for the user" and "use for training" have been separated into different settings. Regulators and customers increasingly expect AI services to offer exactly this. If you are building your own chatbot, two separate toggles are clearer than one that turns everything on or off at once.
And don't expect full anonymity. Even with training turned off, data sits on Google's servers for some time, and with history off it now stays for a full two months. The switch reduces the risk that your text ends up in a future model version, but it does not replace basic hygiene: don't put passwords, keys, or sensitive personal data into the chat.
- Nov 21, 2026Anthropic вышла на биржу Nasdaq: рост выручки за квартал упал со 144% до 47%
- Nov 21, 2026Amazon Plans a 'World Product Graph' — a Database of Every Product on the Internet
- Nov 21, 2026Anthropic shares opened at $111.50, valuing the seven co-founders' stakes at $180 billion
- Nov 21, 2026Applied Digital Asks Oracle and Meta for Billions in Power Guarantees
Key Questions Answered
What changed in the settings?
Previously the choice was rigid. To stop dialogues and attachments from being used to train Gemini, you had to turn off history entirely. It's like not wanting others to read your notebook, so you throw it away after every entry. Now the two toggles work separately. Chat history still supports a configurable retention period, and model training can be turned off without losing access to earlier conversations. You can find the right switch in Gemini's settings under the Activity section.
What falls outside the summary?
The main caveat: training works on an opt-out basis. Google does not ask permission in advance, and until you find the toggle yourself, your chats may end up in training datasets. Opting out applies only to future conversations. Everything you have already sent to Gemini remains in the training data: prompts, model responses, uploaded files, tool calls, location, and data from connected apps. Chats that have already been reviewed and included in a training set are not removed retroactively.
Why storage will grow?
Today, with history turned off, Google still stores data for 72 hours, while the user can access it for 24 hours. After the update, the period will become two months. So access to the data will last longer, both for you and for Google. The company explains this as necessary to respond to you, process feedback, and protect Google, users, and society. In practice, this means Google will be better able to detect AI misuse even if a person has turned history off.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.