Google Saves Search Images and Audio for AI Use

The Verge reports that Google will begin retaining the images, files, audio, and video users supply to Search under a newly named Search Services History setting, according to an email Google sent to users, as reported by The Verge. The Verge lists specific items that will be included: Google Lens photos, recordings from Search Live, voice searches, and phrases spoken into Google Translate. The Verge also reports that Google told users it will use Search Services History to "provide, develop, and improve its services," explicitly including its AI models, and to offer personalized suggestions and ads if personalization is enabled. The Verge says users can turn off Search Services History and related toggles; it also reports that existing Web & App Activity blocks will be preserved during the transition over the "next few months."

Per a May 20, 2025 Google blog post, Google has been expanding multimodal features in Search, including AI Overviews and a newer AI Mode that the company describes as using more advanced reasoning and multimodal inputs. The Google blog explains that AI Mode uses techniques such as query fan-out to break questions into subtopics and issue many queries in parallel. Reporting by The Verge ties the new Search Services History setting to that broader multimodal trajectory by noting the types of visual and audio inputs now covered.

• •Rollout details and timing: The Verge reports a rollout "over the next few months"; observers should confirm the exact activation dates and default states for different regions.
• •Documentation: whether Google publishes detailed retention periods, data-use categories (training, evaluation, personalization), and redaction options in product privacy pages.
• •Regulatory responses: privacy regulators in the EU and other jurisdictions may scrutinize any expansion in the categories of personal data used for model development.

Data provenance and user-consent surface as recurring tensions when large-scale consumer products broaden the kinds of inputs they collect. Reporting from Mashable in 2023 documented an earlier privacy-policy change in which Google stated it may use publicly available information to help train AI products. Together, those reported shifts, the 2023 privacy-policy language and the recent Verge report on Search Services History, are consistent with an industry pattern where product teams consolidate multimodal signals to support AI features that rely on diverse training and evaluation data.

For practitioners focused on data governance, the reported change sharpens questions about dataset documentation, retention windows, and consent flows. The Verge reports that the new setting will be separate from existing Web & App Activity controls and that users can opt out; however, the structure of toggles and the default state during rollout will determine how many users' inputs become available for service improvement. Organizations building downstream models or studies that rely on user-collected datasets should watch how Google documents retention, labeling, and opt-out mechanics.

The Verge's reporting describes an observable expansion in the types of Search inputs Google intends to retain and to use for service improvement and AI development. For data scientists and ML engineers, the key near-term impacts are on dataset availability, user-consent signals, and the provenance documentation needed to absorb such signals responsibly into model development pipelines.