Google has updated the privacy settings for its Search-related services so photos run through Google Lens, audio from voice queries, and files uploaded to Google Translate are now retained by default and used to train Google’s AI models, under new settings called Search Services History and Search Services Personalization. Google Photos is explicitly excluded from the change. The update, described in a June email to users, split what had been a single “Web & App Activity” control into two separate settings, meaning someone who already opted out under the old setting is not automatically opted out of the new one. YourNewsClub frames the restructuring itself as the more consequential change: turning one privacy toggle into two means anyone who set their preferences once and never revisited them is now, without taking any new action, opted in.
The affected surface area extends well beyond Search itself. The policy applies across Google Search, Maps, Shopping, Flights, Hotels, Translate, and News, and covers media people actively submit – a photo snapped for a visual search, a spoken question through Search Live, an audio clip fed into Translate. Saved data can be set to auto-delete after three, eighteen, or thirty-six months, but until deletion, Google’s own language allows that data to be used to develop and improve its AI models. YourNewsClub isolates Google’s decision to exclude Google Photos while including Lens and Translate uploads as the detail worth sitting with: it draws a boundary between media a person deliberately stores and media a person is simply passing through a tool to get an answer.
Opting out currently requires two separate settings changes rather than one – unchecking “Save Media” under Search Services History, and separately reviewing Search Services Personalization to limit how long saved data is retained. Access to these settings is also uneven: Google has said the rollout will continue over the coming months, meaning a large share of the roughly two billion people who use Google Search monthly won’t yet see the option to opt out even if they go looking for it.
The regulatory backdrop makes the design choice harder to read as incidental. Consumer-protection regulators have stepped up scrutiny of default-on AI training practices across the industry, and opt-out complexity – how many clicks, how many settings pages, how discoverable the option actually is – has become a recurring point of concern in that scrutiny. Google has not issued a specific statement addressing whether opting out fully halts retroactive use of media already saved under the previous settings, leaving open whether opting out today stops future training on old uploads, stops only new ones, or both. YourNewsClub flags the retroactive-data question as the gap most likely to matter in practice: for someone who used Google Lens or voice search for months before this update surfaced, opting out now may or may not remove already-collected data from Google’s training pipeline, and the company hasn’t said which.
Owen Radner, who models digital infrastructure as energy-information transport systems, adds a structural read: “Search-adjacent tools like Lens and Translate function as data-collection endpoints regardless of what a company calls them. Framing this as a Search-services update rather than a data-collection policy change is itself a design choice, and it’s one that shapes how many people recognize the update touches them at all.”
Maya Renn, whose work focuses on the ethics of computation and access to power through technology, frames the consent question directly: “Calling this an update that gives users ‘more control’ while defaulting them into expanded AI training inverts what control normally means. A genuine expansion of control would default to the status quo and let people opt in to something new, not default them into the new thing and ask them to find their way back out.” Jessica Larn, who studies macro-level technology policy and infrastructure impact of AI, places the data-scarcity angle: “Every major AI lab is racing to secure real-world data that isn’t already scraped and exhausted from the open web. Search uploads – images, spoken questions, documents people are actively working with – are exactly the kind of high-value, high-context data that text scraped from the internet doesn’t provide. Regulators are still working out whether a buried settings change constitutes meaningful consent for that kind of collection.”
The change fits a broader industry pattern rather than a singular Google decision. Meta has been training its AI on images and other media its users upload, as well as on content captured through its AI-enabled glasses; OpenAI’s consumer terms enable data sharing by default subject to opt-out; and Anthropic’s approach instead requires people to actively opt in before their conversations are used to improve its models. Your News Club calls the difference between opt-in and opt-out defaults the single variable most predictive of how much training data a company ultimately collects, since the overwhelming majority of users never change a default in either direction.