If you use Google, you’re training its AI. Here’s how to opt out.

If you use Google, you’re training its AI. Here’s how to opt out.

Google is rolling out a shift in its privacy settings that allows the company to store and use expanded categories of user data to train its artificial intelligence models. The update, which began rolling out over the next few months, enables the collection of media including images, files, audio, and video recordings. For most users, these settings are enabled by default.

The change was introduced through a "Search Services History" section within Google account settings. According to a customer email sent on June 23, the company framed the update as a way to give users even more control over saved history, citing benefits such as the ability to revisit past visual searches in Lens or continue a Search Live conversation. However, the same communication noted that your saved media is also used to develop and improve Google services and technologies, including AI models and safety measures.

What data is being collected?

The expanded collection scope extends beyond standard text searches to multimodal data. According to Google's help documentation, the company uses history to provide, develop, and improve its services (such as training generative AI models) with the assistance of human reviewers.

Affected services include:

• Google Search and Search Live: Voice inputs and audio recordings.
• Google Lens: Images snapped or uploaded for visual searches.
• Google Translate: Audio from speaking practice.
• Other Services: Maps, Shopping, Flights, Hotels, and News.

This shift reflects a wider industry trend where companies move away from solely scraping the web toward collecting data generated by users. Meta has similarly trained AI on user images, media, and content recorded via its AI glasses. Other competitors follow various patterns: OpenAI's consumer accounts have data sharing enabled by default, while Anthropic uses an opt-in request for Claude to access chats and coding sessions.

How to opt out of Search AI training

Users can stop their media from being used for AI training by visiting the Google Activity Controls or "My Activity" page. If the "Search Services History" tab is available, users can uncheck the Save Media box. This allows users to keep their search history for personalization while preventing media files from being used for AI training.

Alternatively, users can disable the entire "Search Services History" section or set data to be automatically deleted after three, 18, or 36 months. Thorin Klosowski, a senior security and privacy activist at the Electronic Frontier Foundation, argues that companies should use an "opt in" model instead, stating that asking users to consciously enable such features is the least they can do.

A critical detail regarding the timing of these changes is the retention period. A pop-up notification revealed that if saved media is used to train AI models, it is disconnected from the Google Account, but the training data is kept for up to 4 years, even if the original activity is deleted.

The Gmail Controversy

Separate from search services, concerns have emerged regarding Gmail. Cybersecurity experts and a proposed class-action lawsuit allege that Google has automatically opted users in to allow access to private messages and attachments to train AI. The lawsuit claims the company secretly used Gemini to access the recorded history of private communications.

Google has contested these claims. A spokesperson told HuffPost that reports are misleading and stated, we do not use your Gmail content for training our Gemini AI model, noting that "Smart Features" have existed for years.

For those wishing to disable these features regardless, users must opt out in two locations:

1. General Settings: Unselect Smart features.
2. Manage Workplace smart feature settings: Toggle off features for Google Workspace and other products.

Opting out of these settings removes several functionalities, including the "Ask Gemini" summarization tool, "smart compose," automatic email filtering into promotional and social inboxes, and spell-check. It also disables personalized events from emails being added to calendars and restaurant reservations appearing in Maps.

Analysis of the Shift

The move toward multimodal data collection is driven by the need for high-quality training materials that are harder to scrape from public websites than text. Ben Winters, director of AI and privacy at the Consumer Federation of America, suggests that placing the burden of opting out on the user may lead to user exhaustion bordering on nihilism and a feeling of powerlessness regarding data protection.

Google maintains that it uses filters to automatically remove a broad range of identifying info or sensitive personal information and that data is not associated with a user's identity once used for AI training.