Gemini
Google · United States
What it is
Google’s AI assistant and model family. It is a standalone app (web, Android and iOS) and is also built into Google Search, the Workspace apps (Docs, Gmail, Sheets, Meet), Chrome and Android. It drafts, summarises, analyses documents and images, and answers questions grounded in Google Search.
How it works
Gemini is built on Google’s Gemini family of large language models, developed by Google DeepMind. (The version names move quickly, so we describe the family rather than a release.) It grounds answers in Google Search and, in Workspace, in the documents and mail you already have access to. As with the others, how your data is handled depends on which Gemini you are using far more than on the model.
Who is behind it
- Owner
- Google LLC, part of Alphabet Inc.; the models are built by Google DeepMind. CEO Sundar Pichai; DeepMind CEO Demis Hassabis.
- Founded
- Google was founded in 1998; Alphabet was formed in 2015. The Gemini line (formerly Bard) dates from 2023.
- Headquarters
- Mountain View, California, United States. The Google Terms of Service are governed by California law (with a local-courts fallback for some non-US users).
Plans
- Gemini Apps — Free and Google AI Plus / Pro / Ultra · Consumer (free / paid)The consumer assistant; free and paid alike train and human-review by default unless you turn Keep Activity off.
- Gemini in Google Workspace · Enterprise / WorkspaceCore service in Business and Enterprise editions; no training, not human-reviewed.
- Vertex AI / Gemini API (paid) · API directThe developer/cloud layer; no training on inputs. Supports geographic data residency and localized ML processing across global regions, including Australia. NB: the free Gemini API (AI Studio) does train on data.
How it handles your data
The dividing line is consumer versus Workspace/Cloud, and it is sharper than most people expect. Consumer Gemini Apps — the free tier AND the paid Google AI Plus, Pro and Ultra plans alike — train on your conversations by default and are subject to human review; human-reviewed conversations can be kept for up to three years, disconnected from your account, even after you delete the chat. The opt-out is to turn Keep Activity off (which stops future use for training and review, but does not pull back chats already reviewed). The free Gemini API in Google AI Studio likewise trains on and human-reviews inputs. By contrast, Gemini in Google Workspace (a core service), Vertex AI and the paid Gemini API do not train on your content, are not human-reviewed for training, and add enterprise controls and certifications. On residency, data stored at rest remains in your selected cloud location. For enterprise tiers (Vertex AI), machine learning processing can now be locked to specific regional boundaries — including the United States, the EU, and Australia (australia-southeast1). Note that while there is still no local New Zealand processing region, Australian endpoints allow Trans-Tasman data compliance.
Sources for this description
- Gemini Apps Privacy Hub — Googlelive sourcePerma.cc recordas at 31 May 2026
- Manage & delete your Gemini Apps activity (Keep Activity / retention) — Googlelive sourcePerma.cc recordas at 31 May 2026
- Generative AI — Security, Compliance & Privacy (Workspace) — Googlelive sourcePerma.cc recordas at 31 May 2026
- Generative AI in Google Workspace Privacy Hub — Googlelive sourcePerma.cc recordas at 31 May 2026
- Vertex AI — data governance — Googlelive sourcePerma.cc recordas at 31 May 2026
- Gemini API Additional Terms of Service (free trains, paid does not) — Googlelive sourcePerma.cc recordas at 31 May 2026
- Google Privacy Policy — Googlelive sourcePerma.cc recordas at 31 May 2026
- About Google / Alphabet (company facts) — Alphabetlive sourceas at 31 May 2026
What is safe, on which plan
Each cell summarises what the provider’s terms and the applicable rules say. The colour is our summary of the sources, not advice — select a cell to read the reasoning and follow the sources yourself.
- Green
- Amber
- Red
- Black
- Not yet assessed
| Data category | Consumer (free / paid) | Enterprise / Workspace | API direct |
|---|---|---|---|
| 1. Public information | |||
| 2. Internal non-confidential | |||
| 3. Internal confidential (non-privileged) | |||
| 4. Legally privileged | |||
| 5. Personal information (Privacy Act 2020) | |||
| 6. Sensitive personal information | |||
| 7. Children’s information | |||
| 8. Court-protected material |
1. Public information
- Consumer (free / paid)
- Enterprise / Workspace
- API direct
2. Internal non-confidential
- Consumer (free / paid)
- Enterprise / Workspace
- API direct
3. Internal confidential (non-privileged)
- Consumer (free / paid)
- Enterprise / Workspace
- API direct
4. Legally privileged
- Consumer (free / paid)
- Enterprise / Workspace
- API direct
5. Personal information (Privacy Act 2020)
- Consumer (free / paid)
- Enterprise / Workspace
- API direct
6. Sensitive personal information
- Consumer (free / paid)
- Enterprise / Workspace
- API direct
7. Children’s information
- Consumer (free / paid)
- Enterprise / Workspace
- API direct
8. Court-protected material
- Consumer (free / paid)
- Enterprise / Workspace
- API direct
Select any cell to see what the sources say and why.
Statements about provider terms, plans, and defaults will go stale. This profile names its last review date above.
Where to next
Our framework
The reference behind every answer: the eight kinds of information, the five plans, and what each colour means.
Our frameworkGuidance
Our view on the four questions to ask before you use a tool — confidentiality, privilege, privacy, and the professional rules.
GuidanceAbout us
Who is behind this, how we keep it reliable, and why you can trust it.
About us