Google now says human reviewers, including contractors, may see excerpts of personal Gmail messages and Drive files that Gemini pulls in through its Personal Intelligence feature, sometimes the whole item. In 2023 it promised the opposite. The bigger catch: on a personal account, you can't connect Gmail to Gemini without turning on the setting that allows this review.
Google made the change quietly. A note dated October 16, 2026 now sits in the middle of a September 2023 blog post. That post launched Bard Extensions and told users that if they connected Workspace apps, "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 there. The new note says that when users turn on newer features that link Workspace apps to other Google services, that data "can be used to train generative AI models" depending on their settings, and "can also be subject to human review in limited circumstances." Google added the same note to a second 2023 post, on its approach to generative AI guardrails, which had made the same promise.
Gemini's current help page, About personalization with Connected Apps, spells out the details. It drew wide attention this week through a heavily discussed Hacker News thread linking to a GitHub write-up about it. What it describes is a real change for anyone who has linked a personal Google account to Gemini. To see why, you have to follow where an email goes once Gemini reads it.
What Google's documents now say
Personal Intelligence is Google's name for letting Gemini draw on your own Google data. That covers Gmail, Drive, Docs, Calendar and the rest of Workspace, plus Photos, YouTube, Search history, Contacts and Google Wallet. According to Google's product page, it is a beta for personal accounts held by users over 18. It is not offered on Workspace business, enterprise or education accounts. It is also not available in the European Economic Area, Switzerland, the UK or Nigeria.
The help page makes four claims that matter here.
- Human review happens. "When connected to Gemini, a subset of Connected Apps data processed by Gemini is reviewed by human reviewers (including trained service providers) to debug our services, help keep them safe, and improve them."
- What gets used can be the whole message. Google says Gemini does not train "directly" on your inbox or Drive. Instead it uses "summaries, excerpts, generated media, and inferences" from relevant emails and files, and trains on those. It then adds: "where the email or file is brief or particularly relevant to respond to the prompt, the summary may just be the email or file."
- Google tells you not to connect confidential material. "Please don't connect your apps to Gemini Apps if they contain confidential information that you wouldn't want a reviewer to see."
- Feedback widens the review. "If you submitted feedback, or we need to address abuse or harm, human reviewers may also review some of your Gmail inbox or Photos library media."
Google describes safeguards too. Before trained service providers see interactions personalized with Connected Apps, it says it disconnects them from your Google Account and reduces the personal information in them, except where it is dealing with abuse or harm. The page does not say how large the "subset" is or how many reviewers there are. It also doesn't say what "reducing personal information" removes. VibeCodAI found no public statement from Google beyond these documents.
How an email ends up in front of a reviewer
The mechanism is simpler than the headlines suggest. Gemini doesn't send your mailbox to reviewers. When you ask a question such as "plan restaurants for my upcoming trip", which is Google's own example, Gemini searches your connected apps. It pulls out what looks relevant, like a hotel confirmation or a flight itinerary, and places that text into the conversation so it can answer.
Once the text is in the conversation, it is treated like any other Gemini chat. Under the Gemini Apps Privacy Hub, a subset of chats is reviewed by people who judge whether answers were "low-quality, inaccurate, or harmful" and write better ones for training. So the 2023 promise and the 2026 reality differ in one key way. Gmail content that Gemini quotes into a chat is no longer walled off from the chat review pipeline.
Keep that in mind when you read Google's phrase "summaries, excerpts and inferences". An excerpt of a short email can be the entire email. An inference can also reveal more than the text itself. Google's help page notes that connected data may touch topics "like race, religion, and health". It also says Photos data is used to infer "relationships to people in your photos, and where you've been, including by associating your face with corresponding location data and timestamps."
The setting you can't avoid
This is the detail most guides skip. Google's Privacy Hub says the main switch for review is a setting called Keep Activity. Turning it off stops your future chats from being reviewed to improve Google's services and from being used to train its models.
But the same page says: "At this time, other Connected Apps like Google Workspace are unavailable if you turn off Keep Activity." In practice, if you want Gemini to read your personal Gmail or Drive, you have to leave on the setting that allows training and human review. There is no switch that keeps the Gmail connection and turns off review. Google's protections are the de-identification steps described above, not a choice you control.
What turning things off does, and doesn't, do
Based on Google's own documentation, here is what each control does:
- Disconnecting Google Workspace stops Gemini from pulling new Gmail and Drive content into chats. But the help page says disconnecting "doesn't delete any data from Gemini Apps Activity." Content already pulled into past chats can still be used for product improvement until you delete it.
- Turning off Keep Activity stops future chats from being used for training and improvement review. It also disables the Workspace connection. Chats are still kept with your account for 72 hours. Google says it still uses them "to help protect Google, our users, and the public, including with help from human reviewers."
- Deleting your activity does not reach chats that reviewers have already seen. The Privacy Hub says reviewed chats, along with related data such as language, device type, location and feedback, "are not deleted when you delete your activity" and are kept for up to three years, disconnected from your account.
- Sending feedback is the easiest way to open the door by accident. Even with Keep Activity off, a thumbs-up or thumbs-down sends your feedback, "the last 24 hours of your chats", and any content in them, including data from Connected Apps. Specially trained teams review it, and it is kept for up to three years.
- Temporary chats, Google says, are not used to train its models or to improve Gemini with human reviewers. Google still keeps them for 72 hours and may use them, with human review, to protect its services.
If you want to check your own account, start in Gemini's settings under Personal Intelligence and Connected Apps, then look at Gemini Apps Activity. If you've ever connected Google Workspace, it is worth deleting older chats that touched sensitive email. Remember that deletion won't reach anything already reviewed.
Personal accounts versus paid work accounts
This change doesn't apply to everyone. The Generative AI in Google Workspace Privacy Hub says that for business, education and public-sector customers using the Gemini app, "your chats and uploaded files won't be reviewed by human reviewers or otherwise used to train generative AI models outside of your domain without your permission." Those accounts fall under the organization's Workspace contract.
So the same Gemini app now treats the same email two ways. If it sits in a company Workspace account, contracts keep it out of review. If it sits in a personal Gmail account, it can enter review through Personal Intelligence. Many freelancers, small-business owners and families run real business and medical correspondence through personal Gmail. For them, the account type matters more than the content.
The case for Google, and its limits
There are fair points on Google's side. Personal Intelligence is opt-in. The current terms are disclosed in plain language, including an unusually blunt warning not to connect confidential data. Human review of AI chats is standard across the industry, and it is one of the main ways labs catch harmful or wrong answers. Google also says it removes account links and reduces personal information before contractors see anything.
The weakness is in how the change was made. Google's 2023 sentence was a specific, unconditional promise naming Gmail, Docs and Drive. It was reversed for the new features by a parenthetical note added three years later to the same post, not in any prominent notice to users that VibeCodAI could find. And the bundling is a design choice. Google could offer the Gmail connection without improvement review, as it already does for Workspace customers. It hasn't done that for personal accounts.
What to watch
The open questions are ones only Google can answer. How large is the reviewed "subset"? What does de-identification actually strip from an email before a contractor reads it? Will Personal Intelligence reach the European Economic Area and the UK, where it is currently unavailable and where regulators have previously challenged the human review of voice-assistant recordings? Until Google says more, here is the practical reading of its own documents: if you connect a personal Gmail or Drive account to Gemini, assume parts of it may be read by a person, and only connect what you'd be comfortable with that person seeing.