Score Breakdown by Category
How Google Gemini's terms rate across my evaluation categories.
Compare With Other AI Services
Documentation map: which document controls what
Gemini is not governed by one agreement. At least four overlapping Google documents apply, and they answer different questions. This is the single biggest reason ordinary users do not understand their own rights. Use this map to find the document that actually controls the issue you care about.
| Document | What it controls | What to look for |
|---|---|---|
| Google Terms of Service | The baseline contract: acceptable use, the license you grant Google over your content, account suspension, liability limits, and dispute terms. | The content license clause and the suspension and liability sections. |
| Gemini Apps Additional Terms | The Gemini-specific layer on top of the general ToS: age requirements, what the service is, and how it interacts with the rest of your Google account. | Anything that says it supplements or overrides the general ToS for Gemini. |
| Google Privacy Policy | How Google collects, combines, and uses your data across services, including how Google-account signals can personalize and improve products. | The "how we use information" and cross-service combination language. |
| Gemini Apps Privacy Hub / Notice | The operative detail for Gemini chats specifically: human review, retention windows, and what the activity toggle actually does. | Human-review language, retention periods, and the activity-off carve-outs. |
| Workspace agreement + Cloud DPA (business only) | For Gemini for Workspace, this contract supersedes the consumer terms. Google states it does not train on customer data or apply human review to that data under this agreement, and it sets processor obligations. | Whether your login is actually a managed Workspace account, not personal Gmail. |
⚠ Why the layering matters
The privacy detail that most affects you (human review and retention) lives in the Gemini Privacy Hub, not the headline Terms of Service. People read the ToS, miss the Hub, and assume the activity toggle deletes everything. It does not. When the documents appear to conflict, the more specific Gemini and Workspace documents generally control over the general ToS for that subject.
The Google-ecosystem data-flow risk
Gemini's real difference from a standalone chatbot is not the model. It is that Gemini lives inside the same account that holds your Search history, YouTube activity, Maps location, and Gmail. That changes the privacy calculus, because your AI conversations do not sit in isolation.
🔗 How the data can connect
Under Google's Privacy Policy, information from your Google Account and other Google services can be used to provide, improve, and develop Google products, including Gemini. In practice that means a Gemini answer can be personalized using account-level signals, and your Gemini activity becomes one more stream feeding the broader profile Google already holds. The free chatbot is a front door into an account that knows a great deal about you.
⛔ The exposure most users underestimate
If you would never type something into Google Search because of what it reveals about you, treat the consumer Gemini app the same way. A sensitive prompt is not just a question to a model; with activity on, it can be saved to your account, sampled for human review, and combined with the rest of your Google footprint. For a single integrated profile, that aggregation is the risk, not any one message.
✓ How to break the linkage
Turning off Gemini Apps Activity, setting a short auto-delete window, and adjusting your broader Google activity controls (Web & App Activity, YouTube History, Location History at myactivity.google.com) meaningfully reduce how much flows into the combined profile. For genuinely sensitive work, use a business Workspace tier or a provider whose chats stay out of an ad-and-search ecosystem entirely.
Claude vs ChatGPT vs Gemini: terms side by side
The dimensions that decide real risk for a paying or professional user. Every cell below is my reading of each provider's published default terms under my methodology, not a statement of corporate fact, and terms change often. "Best" marks the option I read as most user-protective on that row. Confirm each provider's live agreement before relying on any single cell.
| Dimension | Google (Gemini) | Anthropic (Claude) | OpenAI (ChatGPT) |
|---|---|---|---|
| Training opt-out (consumer default) | As I read the consumer terms, chats can be saved, sampled for human review, and used to improve models, with an opt-out that exists but is not the default | As I read Anthropic's published terms, paid and API inputs are excluded from training by default and consumer-chat training is opt-in BEST | As I read OpenAI's published terms, API inputs are excluded by default while consumer ChatGPT chats are used unless you opt out |
| Data retention after opt-out | The Privacy Hub describes a short window kept even with activity off (around 72 hours), plus longer retention reserved for safety and abuse | As I read it, a clearer published retention schedule, with zero-retention options described for qualifying tiers BEST | As I read it, defined retention with deletion controls, with some safety-related retention applying |
| Multi-layer terms complexity | In my view the hardest of the three to follow: several overlapping Google documents to read | In my view a cleaner separation of free, paid, and API handling BEST | In my view improved, but still several toggles and policies to navigate |
| Ecosystem lock-in / linkage | Tied to the whole Google account; data can combine across Search, YouTube, Maps, Gmail | As I read it, a standalone product without an ad-and-search ecosystem behind it BEST | Reads as largely standalone; check the enterprise terms for any third-party hosting or processing relationships that apply to your plan |
| Business tier that excludes training | Google states Gemini for Workspace customer data is not trained on or human-reviewed under the Workspace agreement and Cloud DPA BEST | As I read Anthropic's terms, API, Team, and Enterprise inputs are excluded from training by default BEST | As I read OpenAI's terms, API and enterprise inputs are excluded from training by default BEST |
| My fairness score | 48 / Grade C | 72 / Grade B BEST | 55 / Grade C+ |
Scores are my editorial assessment under the published methodology, current as of the version dates shown. Vendor terms change often. Confirm the live agreement before you rely on any single cell. See the full Anthropic review and OpenAI review.
Business Workspace vs consumer Gemini: the controls I tell clients to set
A realistic scenario: a small firm wants to draft and summarize inside Google Workspace. The difference between the consumer app and a Workspace plan is the difference between "my client data may be sampled for review" and "my client data is contractually walled off." Here is the scenario and the controls that actually limit exposure.
📝 Scenario
You run a small professional practice on Google Workspace. You want Gemini to summarize client emails, draft documents in Docs, and clean up spreadsheets. Some of that material is covered by NDAs, and one client is in a regulated industry. You bill the output to clients as your work product.
- Buy and verify the Workspace tier. Use Gemini for Google Workspace under a managed organization account, not a personal Gmail in the free app. Confirm in the Admin console that the users handling client data are on the business plan, because the login is what selects your contract.
- Rely on the Workspace agreement, not the consumer terms. Google states Workspace customer data is governed by the Workspace agreement and Cloud Data Processing Addendum, is not used to train models, and is not shown to human reviewers under that contract. That promise lives in the business agreement, so keep a copy.
- Map NDAs and consent before pasting. A client NDA that bars disclosure to any third-party processor without consent can be breached even on a no-training tier. Check the NDA and get consent where required.
- Keep a documented human review step. The terms put output verification on you. Build a sign-off so no hallucinated citation, number, or clause reaches a client unchecked.
- Disclose AI use where your duties require it. Depending on your profession and your client agreements, you may owe a disclosure or consent obligation before processing client material through an AI tool. Many state privacy and professional-conduct rules increasingly expect this.
⚠ The trap most businesses miss
People assume "I pay Google" means business protection. It does not unless you are actually on a Workspace plan and the relevant account is managed by your organization. A personal Gmail using the free Gemini app is on the consumer terms no matter how much you pay Google for unrelated storage. And even the Workspace promise protects you from Google, not from your own duties to your clients under their NDAs and the privacy laws that cover their data.
Highest-stakes points
Where the real money and real risk sit, color-coded by severity.
⛔ Highest stakes: consumer chats can be human-reviewed and retained
Per the Privacy Hub language, on the free app with activity on a sample of conversations may be read and annotated by trained reviewers, and some data is retained for safety even after you turn activity off. In my view that is the term that turns a sensitive prompt into a real exposure. Treat the consumer app like Search, not like a private notebook.
⚠ Watch: account linkage can cascade
Gemini lives inside your Google account, so a suspension or policy action does not always stay confined to the chat tool. If your business depends on Gmail, Drive, or Workspace, do not build a single point of failure on one account with no independent backups and no fallback provider.
⚠ Watch: you carry hallucination liability
Google disclaims accuracy and tells you to double-check responses. Because Gemini is woven into Search, answers can look authoritative while being wrong. If you act on a bad output, the loss is yours under the terms, which is why a human review step is non-negotiable for professional use.
✓ The good news: strong, real controls exist
Unlike some competitors, Google gives you account-level switches that genuinely reduce exposure: the Gemini activity toggle, auto-delete, Takeout export, and the broader activity controls. They are buried rather than prominent, but they work. The Workspace tier provides a contractual no-training, no-review posture for business data.
Analysis