Gemini Shouldn’t Need Human Review to Read Your Gmail

In 2023 Google promised that Gmail, Docs and Drive content used by its chatbot would not be seen by human reviewers. In 2026 that promise has been rewritten. If you connect your personal Gmail or Drive to Gemini, some of what it pulls in can now be read by people, and there is no setting that keeps the connection while switching the review off. That is the wrong trade to ask of users.

This is a DailyAI analysis. The facts below come from Google’s own pages. The view is ours.

What Google said then, and what it says now

When Bard first linked to Google’s apps on 19 September 2023, the launch post was plain: “your content from Gmail, Docs and Drive is not seen by human reviewers, used by Bard to show you ads or used to train the Bard model.” That sentence is still in the original launch post. Beneath it now sits a note dated 16 October 2026. It says connected Workspace data can be used to train generative AI models, depending on settings, and “can also be subject to human review in limited circumstances.”

The detail is on a help page about connected apps. “A subset of Connected Apps data processed by Gemini is reviewed by human reviewers (including trained service providers),” it says. Those reviewers grade Gemini’s answers and suggest better ones. Google says it does not train “directly” on your inbox or Drive. It trains on summaries, excerpts and inferences drawn from them. Then comes the catch: where an email or file is short or especially relevant, “the summary may just be the email or file.”

The page also carries a warning that tells you a lot: “Please don’t connect your apps to Gemini Apps if they contain confidential information that you wouldn’t want a reviewer to see.” For most people, an email inbox is exactly where confidential information lives.

On Hacker News, a discussion of the help page drew wide attention this week. Google has not published a fresh statement on it.

The setting that does two jobs

Google’s answer to privacy worries is a setting called Keep Activity. Turning it off stops your future chats being used to train models or reviewed to improve Google’s services. But the official privacy hub adds a line most users will never see: “At this time, other Connected Apps like Google Workspace are unavailable if you turn off Keep Activity.”

Want AI to save you hours a week? Our free playbook shows you how.Get the free guide →

So the choice is all or nothing. You can let Gemini read your Gmail and accept that people may see some of it. Or you can keep reviewers out and lose the Gmail feature. There is no middle option.

Three more details make that trade heavier than it looks:

• Chats that reviewers have seen are kept for up to three years, and deleting your activity does not remove them.
• A thumbs-up or thumbs-down sends Google your feedback plus the last 24 hours of your chats, including any data from connected apps. This happens even with Keep Activity off.
• Disconnecting Gmail later does not delete what Gemini has already pulled into past chats.

Google says it strips account details and reduces personal information before outside reviewers see anything, except where it is tackling abuse or harm. That helps. It does not make an email from your doctor or your bank anonymous, because the sensitive part is usually the content itself.

Why this is the wrong design

Human review is a normal part of building AI. Graders catch bad answers, and Google is open about doing it. Our objection is narrower. Review of private Gmail and Drive content should be a separate choice, off by default, not a condition bundled into the feature.

The data at stake is broad. The help page lists emails, files, photos, contacts with their phone numbers and birthdays, and Google Wallet details, including financial accounts linked through services such as Plaid. It notes this may include sensitive topics such as health, race and religion.

Other AI assistants face the same question as they reach deeper into people’s lives. Apple has pushed Siri into past chats and voicemails, and OpenAI has had to fix a flaw in its shared chat links. The more an assistant can read, the more its privacy terms matter.

A fix would be simple. Google could offer one switch for “let Gemini use my Gmail and Drive” and a second for “let people review that content”. It already splits audio and Gemini Live recordings into their own setting, which is off by default. Mail and files deserve the same treatment.

Why it matters: Google is steering Gemini users toward letting it read their inboxes, files and wallets. Until review gets its own switch, anyone who connects Gmail should assume a person might read part of it, and leave truly private mail out.