Google Now Lets Human Reviewers Read Gemini-Linked Gmail Content

Google's updated Gemini privacy documentation now allows human reviewers to read information users share with the assistant — including content pulled from Gmail and Drive — reversing a three-year-old promise the company made when it first connected its AI to Workspace apps.

The reversal matters because it touches the most sensitive data most people own. Inboxes hold personal conversations, financial information and work documents. That's also exactly why email is valuable to an AI assistant: it's the richest single source of context an assistant like Gemini can draw on.

What did Google originally promise?

When Google launched the Workspace extensions for its assistant — then called Bard — it made the commitment explicit. "If you choose to use the Workspace extensions, 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," the company said at the time.

Tech columnist Jason Aten, writing about the change, framed the original promise as a big deal precisely because email is so personal. "I don't imagine anyone really wants Google—or any company—reading their email," Aten wrote.

Human review of AI chatbot conversations is standard practice across the industry. Companies use reviewers to check whether chatbots understand requests and deliver accurate responses. There's no other reliable way to verify the tool is doing its job. That means anything an assistant handles — including email content — falls within what humans would normally see, unless a company deliberately excludes it. Google promised it would exclude it. Now, three years later, it has quietly changed the rules.

What does the new policy actually allow?

Under the updated Gemini privacy documentation, several things have changed:

• Human reviewers can read information users share with Gemini, including information from Gmail and Drive.
• The reviewed material specifically includes summaries, excerpts and other information extracted from users' emails and files.
• In some cases, Google says the review can cover an entire email or document.
• If the "Keep Activity" setting is turned on, that same information can also be used to train Google's AI models.
• Conversations reviewed by humans can be retained for up to three years — whether or not the user deletes their Gemini activity.

The three-year retention window is the sharpest edge of the new terms. Deleting Gemini activity, which users might reasonably assume removes their data from Google's systems, does not remove conversations that human reviewers have already processed.

Why the change is hard to police?

The scope of what reviewers can now access is broad by design. An "excerpt" extracted from an email can range from a trivial fragment to a substantial portion of a message, and Google's language allows for the full document in some cases. Users have no practical way to know which of their conversations were reviewed, what was retained, or for how long beyond the stated three-year maximum.

The original 2021-era promise also covered two other uses: ad targeting and model training. Under the new terms, training is permitted when "Keep Activity" is enabled. That setting is tied to users' broader Google activity controls, which means many accounts likely have it switched on by default from years of using other Google services.

What should users watch next?

The practical takeaway for Gmail users: the promise that separated Workspace content from human review no longer holds, and the controls that matter now are Gemini's activity settings rather than any pledge in a policy document. Whether Google extends human review, shortens the retention window, or faces regulatory pushback over the quiet reversal will determine if this becomes a lasting precedent for how AI assistants handle email.

By Amara Osei