Gemini Advanced sits behind Google One AI Premium at $19.99 a month, and almost every write-up about it focuses on model quality, context windows, and whether it beats ChatGPT. Fair enough. But the reason plenty of people stall before hitting subscribe isn’t benchmarks. It’s the nagging question of what happens to the things they type into the box.
So this is a privacy review, not a features review. I went through Google’s own help documentation, the settings that ship switched on by default, and what it all means in practice for anyone tempted to paste something sensitive into a chat. If you want the broader picture of how the assistant actually performs day to day, there’s a separate look at the Gemini AI app and where it fumbles worth reading first.
The detail that shapes everything: consumer terms, not business terms
Gemini Advanced runs on your personal Google Account, which means the consumer Google Terms of Service and Privacy Policy apply. That is a genuinely different regime from Gemini inside Workspace, where Google offers contractual data processing terms, admin controls, and a commitment not to use your content to train models outside your domain.
Same model family. Very different paperwork. If your employer has Gemini switched on for a Workspace account, using that account for work tasks isn’t a small improvement in protection. It’s the difference between a product you happen to use and a service your company has actually negotiated over.
Gemini Apps Activity and the 18-month default
Every conversation you have with Gemini is stored under something called Gemini Apps Activity, which lives in your Google Account alongside Web & App Activity. It’s on by default. The default auto-delete window is 18 months, and you can shorten it to 3 months, stretch it to 36, or turn it off completely.
That one dropdown matters more than any other privacy control in the product. A saved chat isn’t just a transcript. It can include the PDF you uploaded, the photo you asked about, the spreadsheet excerpt you pasted in, and whatever the model wrote back. Eighteen months of that builds a fairly detailed picture of your work, your health questions, your clients, and your opinions about your colleagues.
What turning it off actually does
Switch Gemini Apps Activity off and Google says your conversations are deleted after 72 hours instead of being kept for months. They also stop being used to improve the models. That 72-hour window exists for security and abuse detection, not for training.
There’s also Temporary chats, which work the way you’d hope: nothing saved to your history, deleted after 72 hours, no model training. It’s the closest thing to a private mode, and it’s tucked into the chat interface rather than advertised.
Human review is real, and Google says so plainly
Google’s own support page carries a warning that reads like it was written by a lawyer who lost an argument. Don’t enter confidential information into Gemini, it says, or anything you wouldn’t want a reviewer to see or Google to use to improve its products.
That’s the honest version of the deal. A subset of conversations gets read by trained human reviewers, and while Google says reviews are disconnected from your account, the text itself isn’t anonymised. A pasted contract still names the parties. A health question still describes the symptoms. De-identification protects your identity, not your content.
Regulators have paid attention too. Google held back the Gemini app’s EU and UK launch in early 2024 while it engaged with the Irish Data Protection Commission, which had questions about how the assistant handled personal data. The rollout went ahead after changes were made, but the episode is a useful reminder that this is moving ground rather than settled territory.
Extensions are where the surprises live
Gemini Advanced connects to Gmail, Docs, Drive, Calendar, YouTube, Maps, and more. Each extension is opt-in, and each one gives the model read access to a slice of your Google Account so a prompt like “find the invoice from the plumber” can actually work.
The privacy catch isn’t the permission itself. It’s the combination. Ask Gemini to summarise an email thread and that email content lands in a conversation which, by default, sticks around for 18 months. You’ve effectively copied your inbox into a chat log with a different retention policy than your inbox.
Worth auditing the extension list once. Most people have four switched on that they forgot about months ago.
What Gemini Advanced gets right on privacy
- Controls are usable. One toggle, one dropdown, one delete button. No dark patterns and no nagging to keep history on.
- Temporary chats exist without needing a second browser or a burner account.
- No ad targeting. Google says Gemini Apps conversations aren’t used to serve you ads, and there’s no ad-supported tier quietly undermining that promise.
- Deletion works. You can bin an individual chat or wipe everything, and data is encrypted in transit and at rest.
- Training is opt-out, not forced. Plenty of rivals are less generous with that particular switch.
Where it falls short
The default is the problem. Gemini Apps Activity ships switched on and set to 18 months, and the burden of changing that sits entirely with you. A privacy-conscious design would ask once, in plain language, before saving your first conversation.
Personal accounts also get no zero-retention option. Enterprise customers can negotiate retention windows and pull audit logs. You get a dropdown with a three-month floor. Uploaded files follow the retention of the conversation rather than a schedule of their own, so a single sensitive PDF can outlive the reason you uploaded it.
A practical checklist before you paste anything
- Open Gemini Apps Activity and set auto-delete to 3 months, or switch the whole thing off.
- Use Temporary chats for one-off questions, especially anything with a real name in it.
- Keep work tasks on a Workspace account if you’re lucky enough to have one.
- Strip client names, account numbers, and identifying details from documents before uploading. A find-and-replace takes thirty seconds.
- Audit your extensions and disable the ones you’ve forgotten.
- Clear old chats quarterly. Hardly anyone ever opens that page.
- Treat the input box as public: no passwords, no unreleased code, no medical records, no other people’s data.
Does paying change any of it?
No. This needs stating flatly, because the assumption is common. $19.99 a month buys you the Pro models, 2TB of storage, and Workspace integration. It does not buy different retention defaults, an exemption from human review, or a stricter data agreement.
If you’re weighing up whether the paid tier earns its keep against free alternatives, an honest breakdown of what free AI tools really give you for $0 is a more useful starting point than the pricing page.
Who this is fine for
Brainstorming, drafting, learning, debugging code from public libraries, planning a trip, rewriting an email you’re annoyed about. For those jobs the trade is reasonable and the controls are enough. Millions of people use it exactly this way and sleep fine.
Pause if your prompts regularly contain information that belongs to someone else: client files, patient notes, employee records, embargoed source material, financial figures before an announcement. In those cases the question isn’t whether Google is trustworthy. It’s whether you could justify the disclosure if anyone ever asked, and “the retention setting was on 18 months” is a bad answer.
If that’s your situation, it’s worth reading a straight comparison of Claude vs Gemini and which assistant fits how you work, because the two handle conversation retention and account structure differently.
Who should actually pay for Gemini Advanced
Pay if you already live in Google’s ecosystem, want a capable assistant stitched into Docs and Gmail, and can keep sensitive material out of the chat box. The 2TB of storage alone covers a decent chunk of the price for anyone already buying extra Drive space.
Hold off if confidentiality is your main worry. Nothing in the subscription reduces your exposure, and the free tier runs on exactly the same activity settings. Spending the money and expecting better data terms is a misunderstanding of what’s being sold. A broader comparison of which AI assistant earns a place in your workflow will help you find one that matches your risk appetite as well as your tasks.
Gemini Advanced is a genuinely good assistant wrapped in ordinary consumer-grade data terms. The realistic protection is deciding, before you type, what you’re comfortable having stored for eighteen months. Then go into settings and change that dropdown to three months. It takes about ten seconds, and it’s the highest-value privacy decision in the entire product.