News

Google AI can access some content from Gmail and chats. Here's how to opt out

Google's AI can use emails and chats at different levels if users let it. But it's easy to opt out.

by Anna Rascouët-Paz, Published Nov. 21, 2025


Image courtesy of Getty Images


In November 2025, a rumor circulated that Google's latest update had automatically opted in all users to allow its generative artificial intelligence (AI) program to train on private communications such as emails and chats.

For example, a post said Google automatically opted all those who used Gmail — Google's email service — into allowing its generative AI to train on their emails. The post also purported to show people how to opt out by switching off so-called "smart features" (archived):

As of this writing, the post had gained 6.7 million views and 143,000 likes. The claim spread, including on Facebook and Reddit, garnering reactions, comments and reshares in the thousands. Further, Snopes readers searched the website and sent emails seeking to confirm the veracity of the claim.

Snopes found there was truth to the claim that Google's AI accesses private content, though the AI-powered smart features privacy option in Gmail's settings was a long-standing part of how Gmail inboxes function and only one way to protect users' private content from the tech giant's generative AI. We will outline the details below. 

We asked Google for more details. "These reports are misleading – we have not changed anyone's settings," a spokesperson for Google said in an email statement. "Gmail Smart Features have existed for many years, and we do not use your Gmail content for training our Gemini AI model. Lastly, we are always transparent and clear if we make changes to our terms of service and policies."

By "training," generative AI experts mean providing elements for the model's knowledge base. In the statement, the spokesperson said Google does not feed users' private content to Gemini's knowledge base. However, as we will see, AI can, with users' consent, use their private content to improve features or help users to complete complex tasks.

Gmail settings

It is true that smart features across the Google platform — including Gmail, Chat, Google Drive and Google Meet, among others — are AI-powered. Smart features have used Google's generative AI since 2023, but they used non-generative artificial intelligence before then. 

The generative AI is part of a larger project Google calls Gemini, developed to compete against ChatGPT

In order to offer users features like emails summaries, smart replies, smart inbox management, "help me write" or "nudges" to follow up on emails, Google has integrated into Gmail AI that scours a user's email exchanges to learn more about how this user communicates. The reason the system might suggest appropriate responses to a correspondent's electronic message is that Gmail's generative AI reads the received message and is learning how users might respond.

Using access to emails, the system might offer to add an event to the calendar or to track a flight straight from one's inbox. In other words, it is using one's data to learn and become more efficient. Occasionally, the AI might share insights with Google at large. 

Snopes followed the instructions posted in the X post above to manage the smart features setting and found it was checked by default. We did this in one professional inbox and two personal inboxes that used the Google Workspace platform and found it was the same for all three:

(Gmail settings)

However, a web search revealed posted tutorials about how to switch off smart features as early as 2020. AI-powered smart features have been a long-standing part of Gmail, though the advent of generative AI allowed Google to expand them in 2023. We've asked Google for more details about the timeline.

Clicking on "Learn more" in the settings showed details about how Google does this. For example, it handles the data that powers its smart features differently depending on whether the account is professional/academic or personal

The section also outlines more details on Google's legal basis for using this data for users in the European Economic Area, U.K. and Switzerland, where privacy laws are different. Google says that it relies on "your consent" for using "your Workspace Content & Activity." In other words, it says that if the box for smart features is ticked, the user is giving consent to provide access to their emails for the following purposes:

Google adds that another legal basis for accessing this content is the company's and third parties' "legitimate interests," though it is quick to say that there are safeguards to privacy. The goal for allowing AI to access this private information is "To improve our services," the disclaimer says. In other words, the company may use this data to provide better smart features, but also for research and development — that is, users' private data will be helpful to help Google and third parties to create new AI-powered products. 

The end of this section read "Learn more about how we're protecting your Google Workspace data in the era of generative AI," linking to a 2023 post in which Google explains what it does to protect users' data even as it uses AI to mine their private content.

Snopes unchecked smart features and confirmed that doing so switches off all smart features in the inbox. 

Other Google Apps

Smart features also exist for other applications in the Google Workspace platform, including Google Drive, Google Meet and others. To access them, Snopes went to Drive Settings' privacy section and found that there, too, the system had automatically opted us into smart features that require Gemini to access our content. This was true in both the professional account and both personal ones:

(Google Drive settings)

The panel indicated that switching on smart features allowed Gemini to access users' private content for summaries, drafts, key information and other features. 

Users who wish to protect their content from AI can toggle these off. Google announced it made it easier for people to modify their privacy settings regarding smart features in January 2025.

Gemini Deep Research

While the privacy settings above related to the use and improvement of smart features across Gmail and Workspace, Google offered another level of AI that sought access to private content. This was a different part of the Gemini AI project called "Deep Research." 

Gemini Deep Research aims to help users "tackle complex tasks." This, Google argues, may require research into their private content as well as public content on the internet. This part of Gemini is closer to what ChatGPT offers, except Google intends to refine it based on what it already knows of its individual users, but also on the information and documents they exchange privately. 

To make this happen, Google announced on Nov. 5, 2025, that Deep Research could now access people's private information.

For this part, however, Google did not automatically opt in its users. Instead, they can give their consent for the project to access their private information. Snopes followed the prompts to "Try Deep Research at no cost" and went to settings. This panel invited us to switch on Gemini access to our Workspace:

(Google Gemini Deep Research settings)


By Anna Rascouët-Paz

Anna Rascouët-Paz is based in Brooklyn, fluent in numerous languages and specializes in science and economic topics.


Source code