Google quietly flipped a switch. Starting this month, the company defaults to saving images, audio clips and other media from your Search activity. It plans to feed that material into its AI models. The change arrived in emails to users and a new setting called Search Services History. Many never noticed until the data started flowing.
But users can still push back. Visit myactivity.google.com, head to the Search Services History tab and uncheck the box for Save media. Do it soon. Once the data trains a model, Google keeps it for up to four years even if you delete the original activity. The pop-up warning makes that clear.
Short. Direct. And easy to miss if you skip the email.
This latest move fits a pattern. Google already collects vast troves across Gmail, Gemini, Lens and Translate. The new Search policy expands what counts as training fodder. Uploaded photos for reverse image search. Voice recordings from Translate practice. Files tossed into Lens. All now sit ready for AI improvement unless users act.
WIRED reported the details after receiving the company email on June 23. Author Reece Rogers walked through the exact steps and noted the setting arrived pre-checked. Google spokesperson Davis Thompson told WIRED the controls give users more relevant results and let them revisit visual or voice searches. He offered no answer on why the feature defaults to on.
The Electronic Frontier Foundation sees deeper issues. Thorin Klosowski, senior security and privacy activist there, told WIRED that Google occupies a unique position. Its services run so deep in daily life that users have grown comfortable with the data haul. “Google is in a unique spot compared to a lot of the other companies with this,” he said. Inertia works in the company’s favor.
Ben Winters, director of AI and privacy at the Consumer Federation of America, put it another way. The constant need to opt out adds mental load. “It creates this extra layer of math that a consumer has to do about whether they feel comfortable using the tool they’ve been using for a long time,” Winters explained. He described a growing sense of powerlessness. “There’s an increasing feeling of powerlessness and hopelessness about even trying to protect your data, because every little thing is going to be squeezed out of you.”
But the problem stretches beyond individual users. Publishers face their own headaches. In congressional testimony last year, DeepMind vice president Eli Collins acknowledged that Google’s search organization can train on publisher data even after sites opt out of Gemini model training. A Bloomberg recount captured the exchange: “Once you take the Gemini AI model and put it inside the search org, the search org has the ability to train on the data that publishers had opted out of training, correct?” Collins answered simply: “Correct — for use in search.”
Nieman Lab covered the testimony in May 2025. The story highlighted how Google points publishers to robots.txt as the tool for controlling content in Search. Yet that standard offers no fine-grained way to block AI Overviews while keeping normal results. Publishers must block Google entirely or accept their material in generative answers. An internal Google document from August 2024 showed the company removed 80 billion tokens out of 160 billion after opt-outs. The rest stayed in play for search features.
Recent coverage shows the tension persists. Computerworld outlined the new default earlier this month and urged users to act fast on the media setting. Other reports detail similar friction in Gmail. Smart features there scan emails and attachments. Users must disable them in two places — general settings and Workspace controls — or risk feeding Gemini. AppleInsider and Built In published guides in late 2025 and April 2026 that walk through those exact toggles.
Google insists business customer data stays out of training. Consumer data carries different rules. Gemini chats may go to human reviewers. Turning off activity helps, yet some data lingers for safety and response quality up to 72 hours or longer. The company updated its Gemini privacy hub to explain these retention windows.
So what happens if you opt out everywhere? Search loses some memory. Lens results feel less personalized. Voice features forget context. The trade-off feels real. Yet privacy advocates argue the default should run the other direction. Let users choose in, not force them to hunt settings across a dozen products.
And the pace accelerates. Google’s models grow hungrier for diverse data. Text alone falls short. Images, audio, video — all sharpen capabilities in translation, visual search and live conversation tools. With billions of users, even small percentages yield enormous datasets. That scale explains why the company rarely starts from opt-in.
Critics say this approach shifts responsibility. Users already juggle password managers, two-factor apps and cookie banners. Now they track AI training toggles too. The fatigue shows in forums and social posts. Many simply give up.
Publishers watch the same dynamic. Blocking Google risks traffic collapse. Accepting the terms hands content to competitors’ models as well as Google’s. Some have sued. Others experiment with paywalls or limited feeds. No clean answer exists yet.
Google did announce exploration of better controls. In January 2026, Digiday quoted Ron Eden, principal for product management, saying the company studies ways for sites to opt out of generative AI features in Search without breaking core results. The statement stressed simplicity and scale. Whether those controls arrive before more data flows remains uncertain.
For now the steps stay straightforward. Open your Google account. Head to My Activity. Find Search Services History. Disable media saving. Then check Gemini activity settings and turn off chat retention. Visit Gmail, open full settings and uncheck smart features for Chat and Meet. Repeat for Workspace if you use it.
Do these once. Review quarterly. The defaults shift often enough that yesterday’s choice may not hold.
Powerlessness need not be permanent. Awareness helps. Clear instructions spread. And pressure from users and regulators may yet force cleaner defaults. Until then, the burden sits with each account holder. Turn the knobs. Protect what you can. The data already sent stays sent. Future uploads don’t have to follow.
WebProNews is an iEntry Publication