You Were Always Training Google's AI. Now You Know It.
In June 2026, Google sent out an email that I suspect most people skimmed and deleted. I didn't. I read it twice, then sat with it for a while, because it clarified something I'd been thinking about for months: the relationship between what we feed AI systems and what those systems become.
The short version: Google updated its Search services privacy settings. If you use Google Search, Maps, Shopping, Flights, Hotels, Translate, or News, two new settings now apply to you. Search Services History. Personalized Recommendations. The new Search data setting is enabled by default. Your voice recordings from Google Search Live, your Google Lens images, your Translate voice practice sessions. All of it can now be saved and used, as Google puts it, "to develop and improve Google services and technologies, including AI models and safety measures."
You can opt out. There's a separate 'Save Media' box you can uncheck, distinct from the 'Search Services History' toggle. You can also configure auto-deletion at 3 months, 18 months, or 36 months. The controls exist. They're just not on by default.
What Actually Changed
The previous 'Web & App Activity' setting has been split into two. Web & App Activity data is now separate from a new Search data setting. This matters more than it sounds.
If you'd previously configured your Web & App Activity retention, those changes no longer apply to Google Search services data. Search is now its own category, with its own controls, governed by its own defaults. So even if you thought you'd locked things down before June 2026, the update reset the calculus. Everyone started fresh, opted in.
I'm not writing this to call Google evil for doing it. The company was more explicit than usual about what's happening. "Your saved media is also used to develop and improve Google services and technologies, including AI models and safety measures." Direct quote. Not buried in a terms-of-service footnote. The transparency is notable, even if the default isn't.
The Part Worth Sitting With
What interests me isn't the privacy angle in isolation. It's the feedback loop.
Every voice recording you make through Google Search Live, every image you snap with Lens for a visual search, every Translate voice practice session. That data, saved by default, trains future AI models. The AI learns from how you actually use it. Which means the version of these tools that exists next year is partly a product of what you and I fed them today. You never agreed to be a teacher. You were just trying to identify a plant or read a foreign menu.
Meta is running the same logic with its AI glasses, training on images, media, and content captured during normal use. Different surface, identical principle: real human behavior in real contexts makes better training data than synthetic alternatives, so the companies are collecting it from the moments when you're just trying to get something done.
This is how AI improves. That's not a criticism. It's a description of the mechanism. And the mechanism works whether or not users know it's running.
What to Actually Do
Find that June 2026 email from Google. Open your account settings. The Search Services History setting and the Save Media checkbox are separate from each other. Configure them independently based on what you actually want, not what defaulted in.
Check whether your previous Web & App Activity settings covered what you thought. Given the split, they probably don't cover Search anymore. Set auto-deletion to 3 months if you're going to leave anything enabled.
Then decide what you're comfortable with. Not what you assumed when you set things up two years ago. What you want right now, knowing that your Lens photos and Live voice recordings are the specific media at stake.
The Larger Thing
I think about data flows a lot when it comes to AI relationships, what goes in, what gets retained, what shapes the system you're actually interacting with. The Google update is a clean example of something that's been running quietly for years: AI improves because humans use it, and the humans using it are functionally contributing to the next version.
That's not inherently wrong. Models that reflect actual human behavior are more useful than models that don't. But the choice should be yours to make, not a default you wake up inside of.
The controls are there now. So is the information.
Source: Techcrunch