Google rejects claims that Gmail uses your emails to train Gemini AI

Google has denied viral claims that it is using Gmail messages and attachments to train its Gemini AI model, saying recent posts and articles about a policy change are misleading. The controversy began after social media posts and a report from security firm Malwarebytes alleged that Google had updated its policies to let the company use Gmail content for AI training, and that the only way to avoid this was by turning off Gmail’s “smart features,” such as spell check.

Google spokesperson Jenny Thomson told The Verge that the company has not changed any user settings and that the Gmail smart features mentioned in the posts have “existed for many years.” She also emphasized that Google “does not use your Gmail content for training our Gemini AI model.”

Even so, users may want to review their settings. One Verge staff member reported that, despite previously opting out of some smart features, they later found those features turned back on. Earlier in the year, Google updated its personalization controls so people could manage smart features in Google Workspace separately from personalization across other Google products, such as Maps and Wallet.

Gmail’s smart features go beyond spell check. When enabled, they can track orders, highlight useful information in messages, or help automatically add flights from Gmail into your calendar. In Google Workspace, turning on these options comes with a notice saying you agree to let Workspace use your content and activity to personalize your experience.

According to Google, that language refers to tailoring features and recommendations, not to using the content of individual emails or attachments as training data for Gemini. The company maintains that the recent wave of warnings about Gmail being fed directly into its AI models is incorrect.