Google Saves Search Images and Audio for AI Use

Google's Search Services History rollout illustrates a recurring practitioner concern with default-on AI-training settings: the mechanism used to expand what counts as trainable data can quietly outrun the privacy controls people already configured. Per Tech Times and TechCrunch, the update permanently decouples media retention from the long-standing Web & App Activity toggle, so a setting that privacy-conscious users switched off years ago no longer governs whether their Lens photos, Search Live recordings, or Translate audio get saved and used to train Google's AI models. That structural change, not the mere existence of a new setting, is the detail practitioners should track most closely.

Per an email Google sent users and its own support documentation, the new Search Services History setting spans Search, Maps, Shopping, Flights, Hotels, Translate, and News, and can include "your searches, info from sites you visit with Search services, and generative AI responses," plus associated media. The Verge and 9to5Google both report the covered media includes Google Lens photos, Search Live recordings, voice-search audio, and spoken phrases entered into Google Translate, along with any uploaded files or video. A nested "Save Media" subsetting governs images, audio, and video specifically, and can be toggled independently of the broader history setting. Google's support page states plainly that, as described in its privacy policy, it "uses your history to provide, develop, and improve its services (such as training generative AI models)," and that data is "disconnected from your Google Account" before being used for model training or human review.

Before this change, a single Web & App Activity toggle governed most of this behavior; going forward, per Tech Times, that setting and Search Services History operate independently and permanently, meaning turning Web & App Activity off in the future no longer affects whether media gets saved under Search Services History. Google says existing preferences carried over at the moment of transition, so accounts with Web & App Activity on inherited Search Services History in the on position, and vice versa, but the two settings diverge from that point forward. TechCrunch independently confirmed the same mechanics and characterized the June rollout, announced quietly via customer email, as under the radar relative to its practical scope.

Per Tech Times, this lands against a documented history of disputes over whether Google's privacy controls do what users believe: a federal jury in San Francisco ordered Google to pay $425.7 million in September 2025 to roughly 98 million users after finding Google kept collecting data from third-party apps even after Web & App Activity was switched off (reported by Bloomberg Law); Google separately paid $68 million in January 2026 to settle claims that Google Assistant recorded users without consent (Fox Business); and Google settled with Texas for $1.4 billion in 2025 over unauthorized location, incognito-browsing, and voice-data tracking. None of these cases concern Search Services History specifically, but they establish a pattern practitioners should weigh when assessing Google's current anonymization claims.

Calli Schroeder, senior counsel at the Electronic Privacy Information Center, told HuffPost (per Tech Times) that voice recordings and facial images saved through Lens and voice search function as biometric data regardless of account-level anonymization, and advised users uncomfortable with that exposure to opt out. A Google spokesperson called the biometric framing "speculative commentary" and "inaccurate," stating Search Services History "is not collected for biometric purposes." Independent research lends some support to Schroeder's caution: a 2019 Nature Communications study found 99.98% of Americans could be re-identified in supposedly anonymized datasets using just 15 demographic attributes, a relevant data point given Google says training media already selected for AI training is retained for up to four years even after the source activity is deleted.

Data governance and ML teams should treat this as a live case study in consent-flow design: check both the Search Services History and Save Media toggles directly at myactivity.google.com rather than assuming Web & App Activity settings still apply, document that Google's four-year retention window for already-selected training media survives deletion of the source history, and watch the FTC's stated position that quietly changing terms of service to enable AI training without clear affirmative consent may raise Section 5 concerns. Teams building on Google Cloud AI or evaluating Google's consumer AI data practices should also track whether regulators or further litigation extend the existing Web & App Activity and Assistant-recording precedents to this specific setting.