Google’s latest privacy shift means your uploads can help train its AI — here’s how to limit it

Google has quietly broadened the way it can use people’s activity across Search and its adjacent services, giving the company a larger pool of data to improve its artificial intelligence systems. The change is not limited to text queries. Images, files, and audio or video recordings created or uploaded through Google’s products can now be retained and used for model training unless users change the default settings.

The update arrived through a privacy-settings revision that Google disclosed to customers in June, but it did not attract the kind of attention usually reserved for a major product launch. Instead, it was framed as a way to give people “more control” over saved history and personalized recommendations. In practice, the changes also separate different kinds of data retention, making it easier for Google to continue using certain information even if users adjust older privacy controls.

For anyone who searches the web, uses Google Lens, talks to Search Live, translates speech, or relies on Maps, Shopping, Flights, Hotels, Translate, or News, the implications are worth understanding. The company is not just learning from the information it scrapes from the open web. It is increasingly learning from what people directly create, upload, and say while using Google services.

What changed in Google’s privacy settings

The most important shift is that Google introduced new settings tied to Search Services History and Search Services Personalization. These settings govern how activity is stored and how that activity is used to tailor the user experience.

That sounds routine, but the scope is wider than many users may expect. The settings now cover more than Search alone and extend to other consumer products inside Google’s ecosystem. As a result, media used in everyday interactions can be kept and used to improve Google’s services, including its AI systems.

Google’s own wording makes that clear. In communications to users, the company says saved media may be used to develop and improve its services and technologies, “including AI models and safety measures.” Its help documentation also says the company uses history to improve services, including by training generative AI models, while also protecting Google, its users, and the public with human review.

Some of the retained material is temporary and exists to make the product function. But Google’s language also indicates that certain saved media can be kept specifically for AI training.

Which Google services are affected

The update reaches beyond standard web search. Google’s broader Search Services umbrella includes a range of tools many people use without thinking of them as separate privacy environments.

Search, Lens, and visual queries

If you use Google Lens to identify objects by taking a picture, that image may now be stored in a way that contributes to model improvement. Visual search is increasingly central to Google’s products, and the company appears to be treating those images as valuable training material.

Voice search and Search Live

The same logic applies to voice-based services. If you use Google Search by speaking into the app, or if you use Search Live, the audio recording may be saved. That matters because spoken prompts often contain more personal detail than typed searches, including names, locations, and contextual information users would never include in a standard query.

Google Translate and language practice

Google Translate is also part of the picture. Users who practice speaking through translation tools may be generating audio that can be stored and used in AI development. For people learning a language, that means a feature designed for practice can also become a source of data retention.

Maps, Shopping, Flights, Hotels, News

Google says the policy update applies to related search services as well, including Maps, Shopping, Flights, Hotels, Translate, and News. In other words, the change touches the broader search and discovery layer of Google’s products, not just the familiar search bar on a homepage.

Why this matters now

The update reflects a larger industry trend: companies are no longer satisfied with training AI models only on publicly available web content. They want data that is fresher, more varied, and more directly tied to real user behavior.

That shift has broad consequences. It means the everyday material people generate while using consumer apps is becoming a major asset for AI companies. Photos snapped for a quick visual search, a voice memo converted into search text, or a translation practice session can now be part of the data pipeline that improves future models.

That approach is not unique to Google. Meta has also moved aggressively to use user-created material to improve its AI systems, including images and media shared on its platforms and content captured by its AI glasses. The common pattern across the industry is clear: platform companies are looking inward, mining their own ecosystems for training data rather than relying solely on outside sources.

For users, the stakes are personal. The data being harvested is not always the kind people think of as public. It may include everyday moments, background conversations, private photos, or location-related searches that collectively reveal far more than a single web query ever could.

How Google explains the change

Google presents the update as an expansion of control, not a reduction in privacy. The company says the new settings help users manage what is saved and how long it is retained, while also giving them a way to personalize their Google experience.

Google has told customers that saved media can be used to develop and improve its services and technologies, including AI models and safety measures.

That statement is important because it confirms the company is not merely storing media for short-term product functionality. It is also describing that media as a resource for AI development.

The help pages reinforce the same point. Google says it uses history to improve services and train generative AI models, while also involving human reviewers in safety-related processes. Taken together, the company is describing a system that combines automation, retention, and manual review to refine both product behavior and model performance.

What users can change right now

Although the default may favor data collection, Google does offer settings that let people reduce what is stored. The key is knowing where to look.

Step 1: Review Search Services History

Users can visit the Search Services History page to decide whether Google may save that activity. On this page, the “Save Media” option can be disabled separately from the broader Search Services History setting. That means someone can keep some personalization while opting out of media storage, or disable both.

Step 2: Set auto-delete preferences

Google also allows automatic deletion of saved data after 3, 18, or 36 months. Shortening the retention window can reduce the long-term amount of information available for personalization and training.

Step 3: Check Search Services Personalization

The Search Services Personalization settings let users configure how activity is used to tailor recommendations and the broader Google experience. This can affect which results, suggestions, and related content appear during use.

Step 4: Review broader privacy controls

Google directs users to additional settings pages where they can manage Web & App Activity, Timeline, YouTube History, and related controls. Because the company’s products are interlinked, the privacy impact often extends beyond a single setting page.

• Turn off “Save Media” if you do not want images, audio, or video stored for AI training.
• Review Search Services History independently of Web & App Activity.
• Shorten the auto-delete window to limit retention.
• Check app- and device-level permissions tied to location, microphone, and camera access.

Why the separation of settings matters

One of the most consequential aspects of the update is structural. Google has split what used to be a more unified control scheme into separate buckets.

Before the change, users could adjust saved search data through Web & App Activity settings. Now, that data is divided between Web & App Activity and a new Search data setting, which is turned on by default. That means changing one set of retention options does not automatically affect the other.

This is an important privacy-design issue. Many users assume that reducing historical storage in one place will apply everywhere. The new arrangement makes that assumption less reliable and puts more responsibility on users to check multiple dashboards.

For a company as large as Google, that kind of design can materially shape behavior. Most users will not go deep into settings menus unless a change is widely reported. A default-on approach, combined with separate controls, tends to favor continued data collection.

The bigger AI race behind the policy shift

Google’s update should be viewed in the context of the broader AI arms race. The value of large models depends heavily on data quality, variety, and scale. That has made user interactions inside major consumer platforms extremely valuable.

For years, web scraping was enough to feed much of the industry’s training pipeline. But as AI companies push for better multimodal systems — models that can understand text, images, speech, and video together — the data they need has expanded dramatically.

That is where Google’s ecosystem becomes especially powerful. Few companies sit on as much search, mapping, language, and media interaction data. If those interactions can be retained, reviewed, and used for improvement, they can become a major advantage in training the next generation of AI products.

The tradeoff is obvious. The more capable the models become, the more dependent they are on intimate user data. And the less visible the collection process becomes, the harder it is for users to understand what they are agreeing to.

Consumer privacy and informed consent

The central concern raised by this change is not simply that Google is using data. It is that many people may not realize how broad the use has become, or that the default settings may allow it.

Privacy policy changes that arrive through email and settings-page revisions are easy to miss. They rarely create the sense of urgency that accompanies a public launch or a major breach. Yet for ordinary users, these changes can have a lasting impact on how much of their digital life is stored and analyzed.

That is especially true when a company links privacy controls to the convenience features people already rely on. Search suggestions, faster results, personalized recommendations, and better voice recognition are attractive benefits. But they often depend on the same underlying data collection that also fuels model training.

As a result, the choice is rarely between full privacy and full utility. Instead, users are deciding how much personal data they want to trade for convenience, and whether they are comfortable with that tradeoff being the default rather than an explicit opt-in.

How Google compares with other AI platforms

Google is not alone in using customer data to improve AI systems, but its scale makes the issue more consequential. Meta has taken a similar approach by training AI on user images, social content, and media captured by connected devices such as AI glasses. Other major platforms are also using customer interactions to fine-tune assistant products, search tools, and recommendation systems.

What distinguishes Google is the breadth of its service stack. Search, Maps, email, video, cloud tools, mobile apps, and translation services can all feed into a connected data environment. Even when each product has its own controls, the overall ecosystem can still produce a large and detailed behavioral profile.

That makes transparency especially important. If AI training is happening across different products and settings, users need clear language, not buried notices and fragmented controls.

What users should do next

Anyone concerned about the update should take a few minutes to check their Google privacy dashboard. Even users who are comfortable with some personalization may want to separate convenience features from AI training preferences.

1. Open Google’s Search Services History settings.
2. Review whether “Save Media” is enabled.
3. Check Search Services Personalization and web activity controls.
4. Set a shorter auto-delete period if you want less long-term retention.
5. Inspect camera, microphone, and location permissions on your phone.

People who use Google for work or sensitive personal matters may want to be especially careful about voice and image-based searches. Those formats can expose more context than typed text and may be more difficult to mentally separate from ordinary browsing behavior.

What this says about the future of search

Google’s move is a sign that search is evolving into a far richer data environment than the text-query era ever was. Visual search, conversational input, speech translation, and cross-app personalization all push the company toward a model in which nearly every interaction can become training material.

That may improve product quality. Better multimodal training can lead to more accurate recognition, more helpful assistance, and safer systems. But it also blurs the line between using a service and contributing to its development.

For users, the best response is vigilance. These settings are not frozen in time, and privacy defaults can change without much fanfare. Anyone who depends on Google’s services should periodically revisit the account controls rather than assuming old preferences still apply.

The lesson from this latest change is straightforward: if you use Google’s search ecosystem, you may be part of the data that improves its AI unless you decide otherwise. The controls exist, but they are now spread across several pages, with some features enabled by default. In the age of AI, that is exactly the kind of detail users cannot afford to miss.

For now, the practical takeaway is simple: Google’s AI ambitions are increasingly tied to the data people create while using its products, and users who want to limit that link will need to actively adjust the settings themselves.