Google Says Gemini Users Can Dodge AI Training, but History-Off Chats Will Stay Two Months

Google says new Gemini settings will let you keep chat history while opting out of AI training, but turning history off will soon leave your chats with Google for about two months instead of 72 hours. The training opt-out only covers future chats.

Nov 17, 2026
Google Says Gemini Users Can Dodge AI Training, but History-Off Chats Will Stay Two Months
A woman stares intently at her phone under cool blue light

Google is finally untangling two things it never should have tied together. In a short post dated November 16, the company said new Gemini settings will let you keep your chat history while opting out of generative AI model training. That is a real win.

Read the second paragraph, though. The same post says that if you turn the new History setting off, future chats will be saved for around two months. Until now, according to an archived version of Google's Gemini Apps Privacy Hub cited by TechCrunch, that window was 72 hours. So the switch that used to mean "keep almost nothing" is about to mean Google holds your chats roughly 20 times longer.

Google says both changes will roll out globally in the Gemini app in the coming weeks.

Here is the old trap. Gemini had a single setting, Keep Activity, that governed both history and training. Ars Technica reports that the only way to stop your chats, attachments included, from feeding Google's training was to turn history off, which meant you could never go back to an earlier conversation. The new layout splits them into separate toggles, and Ars says you will find the training switch under Activity in Gemini settings, or through your Google account under Data and Privacy, My Activity, Other activity, Gemini Apps Activity.

It is still an opt-out. Defaults depend on what you chose before. TechCrunch reports that people who had already disabled Keep Activity will start with both new settings off, while people who had it on will keep their history and can separately decide whether Google trains on their chats. In other words, if you never touched the setting, training stays on until you go find the switch.

The opt-out only looks forward

This is the part Google's post does not spell out. Ars Technica reports that the training opt-out applies only to future interactions. Anything you already fed Gemini, which Ars says can include your prompts, its responses, uploaded files, tool calls, location and data from connected apps, is already in the pipeline. Reviewed chats that became training data will not be deleted retroactively, and Ars says Google may keep that material for up to three years.

Then there is the two months. Google frames it as a safety move and says the retained chats serve three purposes.

"They're used to respond to you, process your feedback, and protect Google, its users, and the public." Google

That last clause is doing a lot of work. TechCrunch says Google would not answer questions about what prompted the change or whether it affects how conversations are accessed or reviewed, and simply pointed back to the blog post. TechCrunch adds that the updated Privacy Hub also cites potential system failures as a reason for keeping chats, and says temporary conversations are not used to train models. Ars reads it plainly: Google will be better able to spot nefarious use even when someone disables history.

That may be a fair goal. It also puts Google out of step with its rivals. TechCrunch notes that OpenAI says it may keep temporary ChatGPT chats for up to 30 days, and Anthropic keeps incognito Claude chats for 30 days by default, though both can hold some data longer for legal or safety reasons. Google's new floor is about double that. Compare it with Proton's pitch for Lumo, which claims even Proton cannot read your chats, and the gap in philosophy is obvious.

Google also shipped temporary threads for AI Mode in Search on November 17, a separate feature. The company says those sessions are completely detached from your account and that once you close one, it is gone. Google has not said whether those threads fall under the two-month Gemini rule.

Net result: more control on paper, and a privacy switch that now keeps more of what you say. Turn off training if you care. Just do not assume turning off history means Google forgets.