Google Says Gemini Doesn't Train on Your Emails — With Exceptions

By James Calloway

Google's Gemini AI can read your Gmail inbox, summarize your messages, and draft your replies — and many of those features are switched on by default. The company insists this does not mean your emails are training fodder. The language in its own policy pages, however, carves out exceptions large enough to deserve a close read.

This year Google rolled out a wave of Gemini-enabled features across Gmail, its email service with well over a billion users. The additions include voice-activated commands, live AI search features, AI-generated summaries of recent emails, automatic to-do list creation, and drafted replies. Paid subscribers get more, such as AI-enhanced proofreading. The productivity pitch is familiar: let the model into your inbox, and it will organize your digital life.

The catch is the price of admission. Many of the free features are enabled by default, which has raised concerns about whether Google's deployment of Gemini across its email suite violates user privacy. The question is not abstract. A wave of lawsuits currently accuses the AI industry of mass copyright infringement, and users can be excused for wondering whether their private correspondence is being treated as raw material for Google's models.

So, is Google feeding your emails to Gemini? The answer is both yes and no.

What does Google actually promise?

Google released a statement in April 2026 detailing how Gemini interacts with user data. It states: "Google does not train its foundational AI models, including Gemini, on your personal emails." Instead, Gemini accesses your inbox only for "isolated tasks," such as summarizing emails or drafting responses.

Google's own framing of the limitation is worth quoting in full: Gemini "only processes your information to complete your specific request, and doesn't hold onto that data afterward." In other words, the model reads your mail to do what you asked, then — according to Google — forgets it. That is a claim about internal data handling that users cannot independently verify.

The promise extends beyond Gmail. Google's support page states the safeguards cover Workspace apps including Chat, Calendar, Docs, Drive, Sheets, and Slides. The exact wording: "We do not use your Workspace data to train or improve the underlying generative AI and large language models that power Gemini, Search, and other systems outside of Workspace without permission."

Read that sentence again. The qualifier "without permission" is doing heavy lifting.

Where do the privacy promises stop?

The permissions change the moment you link Google's content apps across your account. Google notes that when you connect content apps from Google Workspace to Search services — which can enable AI Mode features — it may "use that data to improve Google services for everyone, including by training generative AI models."

Google also says Search services "do not train generative AI models directly on your Gmail inbox." But the very next sentence in its policy concedes the indirect route: "when you interact with Search, summaries, excerpts, generated media, and inferences from your relevant media, emails, and files may be used to help us answer your prompts. To make our responses relevant, helpful and high quality, we train our generative AI models off of these summaries, excerpts, generated media, and inferences."

That is a meaningful distinction, not a loophole-free guarantee. Google says it does not train on your emails themselves — but it does train on the derived outputs produced when your emails feed into Search interactions. Derived data, in Google's framing, is fair game once you have connected those services.

In plain words: the inbox itself is off the training menu, but the AI's summaries and inferences drawn from it are not.

How do you turn Gemini off?

Google does provide controls, though finding them requires a short walkthrough. The settings apply account-wide, so you only need to complete the process once.

In the Gmail app:

• Tap the three lines at the top left corner
• Scroll down and select Settings, then choose your email account
• Go to Smart features and deselect the box
• To remove Gemini from Workspace apps, tap Google Workspace smart features and clear the checkboxes

In a web browser:

• Click the Settings gear icon beside the search bar at the top right
• In the dropdown, select See all settings
• In the General tab, scroll to Smart features, uncheck the box, and confirm
• Return to the General tab, navigate to Google Workspace smart features, click Manage Workspace smart feature settings, turn off both options, and click Save

The controls exist. What Google does not advertise is what opting out costs you.

What do you lose by disabling Gemini?

Turning off smart features can also switch off spelling and grammar checks. Users are placed in an uncomfortable binary: keep the potentially invasive AI capabilities, or give up basic editing assistance and trust their own grammatical acuity.

For users who want to keep the AI features but restrict what Google uses for training, two additional settings matter: Search Services History and Personalized Recommendations. By default, both inherit your most recent Web & App Activity and Search Personalization settings — meaning many users have never consciously chosen them.

• Search Services History: Sign in to your Google account, go to your Search Services History settings, and turn it off. You can also delete previously saved activity. To keep the setting on but limit what is stored, adjust your Save Media settings.
• Personalized Recommendations: Go to your Google account, then Data & privacy > Personalization settings, and toggle Personalized Recommendations in Search services on or off.

Why does this matter beyond Gmail?

The stakes here are larger than one inbox. Big Tech has placed AI at the center of every application, effectively offering customers a Faustian deal: hand over your privacy in exchange for productivity gains — gains that, as the source notes, remain very much debated. Whether or not the efficiency materializes, the data access arrives first, by default, and often before users understand what they have consented to.

Google's April 2026 statement is a useful data point in a broader fight over AI and personal data, and it is more reassuring than most. But the fine print — "without permission," the derived-inference carve-out, the bundling of spell check with AI features — shows how narrowly the privacy promise is drawn. As AI Mode features spread across Google's services, expect the boundary between "processing your data" and "training on your data" to keep shrinking, one permission grant at a time.