Gemini users who switched off chat history expected deletion within 72 hours; Google now holds those conversations for about two months, the French tech blog Korben reported this week.
The change applies even to temporary chats, the mode that never appears in history at all, and it reaches the Gemini app worldwide over the coming weeks.
Google announced the privacy overhaul on November 16, and it affects Swiss users as well. The company is splitting a single Gemini setting into two: one decides whether and how long conversations are saved, the other whether chat content, including attachments, may be used to train its generative models.
On paper, users gain control. The catch sits elsewhere. A person who disabled history precisely so Google would forget their questions now waits roughly twenty times longer for deletion, Korben reported, and the world’s rival assistants forget far faster.
From 72 Hours to Two Months
The old rule tied retention to consent. Anyone who refused to let Gemini chats train Google’s models could not keep them longer than 72 hours, and Google used that window to generate answers and run its safety measures, as t3n reported chats couldn’t outlast 72 hours.
The new window stretches to about two months. Google’s updated Privacy Hub justifies it with the need to answer users, process feedback, and protect users, the service and the public, t3n reported. Google also says chats must remain available if a system fails, and that Gemini can draw on them as context when answering.

That is the official explanation in full. What it doesn’t contain is a reason the same tasks suddenly need weeks instead of days.
One Setting Becomes Two
Until now, saving chats and consenting to AI training were effectively bundled: agreeing to one meant agreeing to the other under the menu item for saving activity. The update severs that link, so a user can keep a searchable history while withholding chat content from model training, Swiss-Press reported.
The defaults depend on what a user had before. Anyone who had kept activity off gets both new settings switched off, Gazeta Express reported. Anyone who had it on keeps their history, and the use of chat content for training stays switched on unless they change it, according to Swiss-Press.
With history enabled, Google stores conversations for up to 18 months, t3n reported. Temporary chats still won’t train the models, Gazeta Express noted. Google Workspace customer data remains excluded from training without consent, and switching off the training setting doesn’t stop Google from processing chats for safety purposes, Swiss-Press reported.
The Feedback Loophole Lasts Three Years
Rating an answer changes everything. Give a thumbs up or thumbs down, and Google can access the last 24 hours of that conversation, attachments included, even with history switched off, Korben reported.
A human reviewer can then read the exchange and its files. Google keeps such reviewed conversations and related data for up to three years, stored separately from the Google account, Swiss-Press reported the three-year retention survives deletion.
Rivals Delete After 30 Days
The competitive contrast is stark. OpenAI stores temporary ChatGPT chats for up to 30 days, and Anthropic holds Claude’s incognito chats for the same period, t3n reported, citing TechCrunch. Google now outlasts both by a wide margin.
| Service and mode | Retention |
|---|---|
| Gemini, history enabled | Up to 18 months |
| Gemini, after human review of feedback | Up to 3 years |
| ChatGPT, temporary chats | Up to 30 days |
| Claude, incognito chats | 30 days |
Both competitors reserve the right to keep some data longer for legal reasons, such as investigating policy violations, t3n reported.
The Question Google Left Open
Why does deleting the same conversations suddenly take twenty times longer? Google didn’t answer when several outlets asked, Korben reported. Pressed on what prompted the change and whether it affects access to or review of conversations, the company referred questions to a blog post instead of answering directly, Gazeta Express reported.
The timing cuts against the feature’s own origin story. Temporary mode launched in 2025 as a promise of ephemerality, a room for embarrassing or sensitive questions that would vanish almost immediately, Korben wrote. Two months is not ephemeral by any ordinary reading of the word.
Korben’s own account adds a practical footnote: the blog’s author disabled Gemini history some time ago, and apparently it didn’t help much. The conversations are still there. They just have a longer lease now, and the deletion clock doesn’t start until the update finishes rolling out to the Gemini app worldwide in the coming weeks.