Google now keeps Gemini temporary chats and chats you send with History turned off for about two months, not 72 hours. The change appears in the Gemini Apps Privacy Hub updated November 16, 2026, and in Google's product blog. Google says longer retention helps respond to you, process feedback, and protect users, including with human reviewers.
For builders and privacy teams, this is a data-retention policy shift, not a model launch. It lands the same week as OpenAI GPT-6.2 and Gemini Tasks expansions, when more people route agentic work through consumer Gemini accounts.
TL;DR: what changed
| Topic | Before (Oct 26 hub wording) | After (Nov 16, 2026 hub) |
|---|---|---|
| Temporary chats | Retained ~72 hours | Retained ~two months |
| History off | Retained ~72 hours | Retained ~two months |
| Used for training? | Temporary chats not used for training | Unchanged: temporary chats not used for training |
| History vs training | Single Keep Activity toggle | Split settings rolling out over coming weeks |
| Official reason | Short operational window | Safety, abuse response, disaster recovery |
TechCrunch's November 17 report cites an archived November 12 Privacy Hub snapshot showing the old 72-hour language.
What Google says it does with those chats
According to the Privacy Hub FAQ "How long does Google retain my temporary chats…", temporary and History-off chats are retained with your account for around two months and used to:
- respond to you, including using the last 24 hours as context;
- process feedback you submit;
- help protect Google, users, and the public, with human reviewers;
- remain available in case of a potential system failure.
Google explicitly states temporary chats are not used to train its AI models. If History is off and you do not submit feedback, Google also does not use future chats to train generative AI models.
That is different from saying the chats are ephemeral. They are not shown in Gemini History, but they still exist on Google's backend for the retention window.
Splitting History from AI training
Checkbox beside an exit door, for separate Gemini History and generative-AI training toggles
Google's November 16 blog introduces separate controls:
- History — whether chats appear in your saved Gemini History.
- Chat content including text and uploads — whether Google may use future chats to train generative AI models.
Previously, one Keep Activity setting bundled both ideas. Google says users who disabled Keep Activity will start with both new settings off by default, while users who enabled Keep Activity keep History on and can decide training separately.
For enterprises, the important comparison is Gemini keep-activity training post and Workspace hubs, where administrators set retention and human review rules.
Implications for developers and teams
Fading sticky note with a pin, for temporary chats that are saved longer on Google's servers
Consumer Gemini in prod workflows
If your team pastes customer data, credentials, or unreleased code into Gemini:
- Temporary chat is not a zero-retention channel anymore — plan for ~60 days server-side retention.
- Feedback on a response can pull 24 hours of context and trigger longer reviewer retention.
- Connected Apps and agentic Tasks may multiply data paths; deleting History does not wipe remote browser or task artifacts per hub footnotes.
Read alongside Perplexity isolated desktops when comparing personal agent trust zones.
Compliance questions customers will ask
| Question | Plain answer |
|---|---|
| Is History off the same as "do not store"? | No. Google still keeps chats ~two months for operations and safety. |
| Is two months a training opt-in? | Training is governed by the new training toggle, not by retention length alone. |
| What about EU users? | Google Ireland Limited serves EEA users; legal bases are listed in the hub EU section. |
| Work/school accounts? | Workspace Generative AI Privacy Hub terms differ; admins control history and training. |
Search and incident response
Security teams should update data maps and vendor assessments to reflect 60-day retention for History-off consumer Gemini, not 72 hours. Incident playbooks that assumed three-day erasure need revision.
How this compares to ChatGPT temporary chat policy
OpenAI's consumer products use different labels — ChatGPT temporary chat gained save-and-personalize options in 2026. This Google move is the opposite direction on retention: longer server-side storage for privacy-oriented modes.
Neither company's marketing matches infra reality without reading help center text. Treat Privacy Hub / OpenAI privacy pages as the contract, not keynote slides.
What to do before the rollout finishes
Google says settings roll out globally over coming weeks. Until you see both toggles:
- Open Gemini History at gemini.google.com and document current Keep Activity state.
- Use Workspace or API projects with billing for workflows that must avoid consumer human review rules.
- Ban pasting secrets into consumer Gemini regardless of History — retention got longer, not shorter.
- Re-read feedback tooltips before clicking thumbs — feedback extends review pipelines.
For regulated teams, pair this policy read with Sweden government ChatGPT blocks as an example of sovereign workplace decisions.
Related reading
- Gemini Tasks free rollout
- Gemini keep activity and human review
- Cloudflare temporary accounts for agents
- GPT-6.2 launch same day
- New York AI testing law context
- AI homework vs exam integrity
Builder takeaway
Google traded short ephemeral retention for operational safety on the same SKUs users pick when they want less history visibility. That is a policy choice product managers should explain plainly to customers — especially anyone who believed History off meant gone in three days.
If your app embeds Gemini through OAuth or Connected Apps, disclose 60-day backend retention in your own privacy notice even when you never store prompts yourself.
FAQ-style scenarios (plain language)
"I turned History off — can Google employees read my chats?"
Google's hub says human reviewers may see data collected for abuse and safety purposes. Retention length and review are not the same as showing chats in your History UI. If your threat model includes state-level adversaries or competitive espionage, consumer Gemini with History off is still cloud processing, now with a longer server window.
"Does Gemini API billing avoid this policy?"
Google AI Studio and paid API projects follow Google API terms with different abuse-logging rules. The two-month rule described here applies to Gemini Apps (gemini.google.com and mobile assistants), not necessarily every API call. Map each integration separately.
"What about deleting History manually?"
Deleting Gemini History does not delete all derived logs instantly. Reviewed feedback, Connected Apps artifacts, and remote browser sessions may persist per the hub's cross-links. Run a tabletop exercise before promising customers instant erasure.
"How does this interact with agent payments?"
Gemini Tasks can place calls, send messages, and spend money. Longer retention increases the window for forensic review if an agent misfires — useful for fraud teams, uncomfortable for whistleblowers.
Timeline: how fast the policy moved
| Date | Event |
|---|---|
| Oct 26, 2026 (archived hub) | Temporary and History-off chats retained 72 hours |
| Nov 12, 2026 | TechCrunch cites archived hub before update |
| Nov 16, 2026 | Google blog announces split settings and longer retention |
| Nov 16, 2026 | Privacy Hub "Last updated" stamp |
| Nov 17, 2026 | Outlets report global rollout over coming weeks |
EU and UK legal bases (high level)
The hub's EU section lists contract performance, legitimate interests, and consent for different processing steps. Privacy teams in the EEA should read the full legal bases FAQ rather than this summary — retention changes can shift DPIA conclusions even when training opt-outs remain.
For educators and exam proctors
Longer cloud retention is the mirror image of South Korea's CSAT anti-cheating sprint: institutions tighten physical exam controls while consumer AI vendors lengthen server-side memory for safety. Academic honor codes should assume weeks of vendor retention, not hours.
Media literacy note
Outlets including The Register framed the shift as 20x longer retention when History is off — arithmetic from 72 hours to ~60 days. The number is directionally fair; the policy trade is safety versus ephemerality, not a hidden training grab, according to Google's own training FAQ lines quoted above.
Document your org's stance in the same place you documented ChatGPT enterprise data policies — one internal wiki beats ten anxious Slack threads.
Checklist for security reviewers
- Confirm whether your workforce uses consumer or Workspace Gemini SKUs.
- Add ~60 days server retention to data-processing agreements for consumer SKUs.
- Re-run secrets scanning training — History off no longer implies three-day deletion.
- Update parental consent flows if minors use Gemini on family accounts.
- Log a ticket to revisit settings after Google completes the split toggle rollout.
Why Google gives for failure recovery
Google explicitly mentions system failure recovery as a reason chats stay available for two months. That is plausible engineering — disaster playbooks need recent prompts to debug outages — but it also means your content rides along in those bundles. If your legal team forbids cross-customer incident bundles, Workspace with negotiated DPAs may still be the cleaner path. Schedule a quarterly reminder to re-open the Privacy Hub — Google updates this page without a keynote or press event, often quietly.
