Google has once again revised its data usage policies, now allowing saved media from search interactions—including images, videos, and voice recordings—to be used for training its large language models (LLMs). The change, announced via email last month, automatically opted all signed-in users into the expanded program. While Google frames this as an improvement to its AI search capabilities, privacy advocates warn that it erodes user control over personal data.
What Does the New Policy Mean?
According to Google’s updated documentation, any media you upload or generate while using Search services (such as photos taken with Circle to Search, voice queries via Search Live, or videos uploaded for analysis) may be saved to your Search Services History. This saved media can then be used to “develop and improve Google’s AI models and technologies.” The policy applies to all signed-in users who have not manually opted out.
The change raises significant privacy concerns. For example, if you use a Google chatbot to discuss sensitive topics like a potential job change, that conversation—including details about your employer, boss, or reasons for leaving—could be used to train the LLM. Later, if your boss asks the same chatbot about employee loyalty, the system might reveal familiar information, effectively doxxing your plans. Similarly, voice recordings used for language practice or search could be repurposed to create synthetic audio without your consent.
Background: Google’s Evolving AI Training Practices
This is not Google’s first expansion of AI training data. In 2023, the company began using public YouTube comments and search queries to train its Bard (now Gemini) model. Earlier this year, it updated its privacy policy to allow scraping of publicly available data from the web. Now, the focus has shifted to personal media uploaded during active search sessions. The trend reflects a broader industry push to feed LLMs with diverse, real-world data—often at the expense of user privacy.
Google’s defenders argue that the data is anonymized and used only to improve AI responses. However, privacy researchers note that anonymization techniques can sometimes be reversed, especially when combined with other data points. Moreover, the automatic opt-in model means many users will remain unaware that their media is being used for training.
How to Opt Out: Step-by-Step Guide
Fortunately, Google provides settings to disable these features. Follow the steps below to prevent your media from being used for AI training.
1. Disable Search Services History
Open the Google app on your mobile device (or visit myactivity.google.com on a desktop). Tap your profile picture in the top-right corner, then select Search history. On this page, ensure the Save media option is unchecked. Next, tap the top entry for Search Services History and toggle it to Off.
2. Turn Off Search Personalization
Return to the Google app, tap your profile picture, and select Search personalization. Tap the top entry and make sure the service is Off.
3. Adjust Data Retention Settings
If you choose to leave any features enabled, you can set automatic deletion periods of 3, 18, or 36 months. This ensures your data is not stored indefinitely. Access these options from the same Search Services History page.
4. Disable Personalized Ads
Go to myadcenter.google.com and turn off Personalized Ads to prevent Google from using your search history for targeted advertising.
Additional Considerations
These settings only affect data collected while signed into your Google account. If you use services like Google Assistant, YouTube, or Google Photos, separate privacy controls may apply. It is recommended to review all permissions periodically, especially after major policy updates.
For those seeking maximum privacy, consider using locally installed AI models (such as those from Hugging Face or Ollama) that do not send data to external servers. This approach eliminates the risk of cloud-based training entirely and gives you full control over your information.
Potential Downsides of Opting Out
Disabling these features will reduce Google’s ability to personalize search results, recommendations, and ads. For instance, you might see less relevant advertisements or lose the convenience of revisiting past visual searches. Some users also appreciate the time-saving aspect of predictive search suggestions. Weigh these benefits against your privacy comfort level.
It is also worth noting that Google may continue to collect aggregated, non-identifiable data for model training even if you opt out of media saving. The company’s privacy policy includes clauses about using anonymized data for research and development.
Industry Context
Google is not alone in this approach. Competitors like OpenAI, Microsoft (via Copilot), and Meta have similar programs that use user interactions to refine their models. However, Google’s extensive ecosystem—spanning search, maps, email, and cloud storage—means the potential data pool is vast. Regulatory bodies in the EU and California are scrutinizing these practices under GDPR and CCPA, but enforcement has been slow. Users in Europe may have stronger protections, including the right to object to data processing.
In response to criticism, Google has emphasized that users always retain control over their data through the My Activity dashboard. The company also claims that media is not used for training if the “Save media” option is unchecked, though this has not been independently verified.
Ultimately, the decision to opt in or out hinges on your trust in Google’s data handling and your personal definition of acceptable privacy. As AI becomes more integrated into everyday search, these trade-offs will only grow more frequent. Staying informed and periodically reviewing your privacy settings is the best defense against unintended data use.
Source: ZDNET News