Google denies claims of Gmail data being used to train Gemini AI: Details

Google said that reports suggesting Gmail emails and attachments are being used to train its Gemini AI are misleading, adding that smart features do not feed model training

Google has issued a clarification after several reports claimed that Gmail was now using users’ emails and attachments to train its Gemini AI models. These reports suggested that Google had changed its policy and that the only way to opt out was by disabling Gmail “smart features” such as spell check or order tracking.

 

However, Google says that this interpretation is incorrect. In a statement to The Verge, Google spokesperson Jenny Thomson said that “these reports are misleading – we have not changed anyone’s settings, Gmail Smart Features have existed for many years, and we do not use your Gmail content for training our Gemini AI model”. 

Google denies Gmail AI training claims

According to Google, the recent confusion largely stems from updated wording and placement of Gmail’s “smart features” settings, which some users saw suddenly surface in their accounts. The settings control whether Gmail can analyse message content to enable features such as categorising emails, detecting spam, suggesting replies or automatically adding travel bookings to Google Calendar.

 

Thomson reiterated that this analysis is used only to power Gmail’s built-in functions and does not feed into Gemini AI model training.

Despite Google’s statement, The Verge notes that at least one of its staff members found that their smart features settings appeared to have been switched back on after previously being turned off, raising questions about how consistently the changes were rolled out.

Why users thought Gmail was training AI

The misunderstanding was amplified by an article from security firm Malwarebytes, which initially claimed that Gmail content might be used to train Google’s AI models and that users were being opted in automatically. The post has since been updated, acknowledging that the publication “contributed to a perfect storm of misunderstanding”.

 

Malwarebytes now states that:

• The settings themselves are not new
• Gmail does scan email content to power smart features
• This processing is separate from AI model training
• Google documentation indicates the setting is opt-in

What Google actually changed

In January, Google introduced more granular controls for Gmail and Workspace users, splitting the original smart features toggle into two separate settings:

• One governing personalisation inside Google Workspace (Gmail, Calendar, Meet, etc.)
• One governing personalisation across other Google products (such as Maps or Wallet)

Google’s blog at the time stated that the update “does not alter our underlying data handling practices” and was intended to give users clearer control over how their data affects product behaviour.

 

These settings determine whether Google can personalise experiences such as:

• Adding flights from Gmail to Calendar
• Surfacing Drive files in Gmail via Gemini
• Compiling tickets and loyalty cards into Wallet

According to Google, none of these permissions allow Gmail content to be used to train Gemini models.