Gemini's data controls, explained
What Google does with what you type — and why the answer is completely different at work and at home.
Updated 31 Jul 2026 fast-moving — check a current source
Behaviour per Google's Workspace privacy documentation, checked late July 2026. Consumer app settings change more often than enterprise commitments — verify before relying on either.
“Does Gemini train on what I type?” has two different answers depending on which Gemini you are using, and the gap between them is wide enough to matter. This is the same shape of question as with ChatGPT and Copilot, with one Google-specific wrinkle.
The enterprise answer
For Gemini in Google Workspace and Gemini for Cloud, Google’s documented position is that enterprise data is not used to train models and is not reviewed by humans. Content stays inside your organisation’s existing access boundaries — Gemini cannot surface a document to someone who could not already open it.
That last point is the one people miss. Gemini respects Workspace permissions, so it does not create new access. But it does make existing access far more discoverable. A file that was technically shared organisation-wide and practically invisible is now something an assistant can find and summarise on request. Over-broad sharing that was harmless through obscurity stops being harmless.
The consumer answer
The consumer Gemini app, signed into with a personal account, runs on consumer terms. Human review of some conversations and use of data for improvement are governed by your account settings rather than an enterprise commitment — and the defaults are not the same as the enterprise defaults.
The riskiest moment is not choosing the wrong setting. It is being signed into a personal Google account on a work machine and not noticing which Gemini you are talking to.
Controls worth knowing about
If you administer Workspace, the relevant levers are:
- Trust rules for Drive sharing — constrain what Gemini can reach by constraining sharing itself, internally and externally.
- DLP and information rights management — where IRM is applied to sensitive files, Gemini does not retrieve them. This is the cleanest way to carve out material that should never reach an assistant.
- Client-side encryption — with CSE enabled, the content is unreadable to Google and to generative features alike. The strongest control, and the one with the most workflow cost.
- Logging and export of Gemini activity — so you can review what was actually asked and accessed.
For most organisations the effective work is not configuring Gemini. It is fixing the sharing sprawl underneath it, which was already a problem and is now a visible one. See shadow AI for the other half of that picture, and Gemini in Workspace for how it behaves once switched on.
Sources
Everything above was checked against these on 31 Jul 2026. Providers change things without notice — if a detail matters to a decision, follow the link.