Apple’s new Siri runs on Google’s AI. That fact has left a lot of iPhone owners asking one simple question: does Google see what you ask Siri? The short answer is that Google is not supposed to see who you are — but a few parts of that promise rest on a contract, not on code.
The timing is why people are talking about it now. John Ternus took over as Apple’s chief executive on September 1, 2026, ending Tim Cook’s 15-year run. Apple’s next product event is on September 9, where the rebuilt Siri is expected to appear on new iPhones. So millions of people are about to start using a Siri that leans on a Google model.
Here is what is confirmed, what is only promised, and what you can actually control.
What actually changed inside Siri
Apple spent two years struggling to ship a smarter Siri. In January 2026, it gave up on doing the whole job alone. Apple agreed to license a custom version of Google Gemini — Google’s family of large AI models — to power Siri’s hardest thinking. Reports across outlets put the price at roughly $1 billion a year, for a custom model of about 1.2 trillion parameters (parameters are the internal settings a model learns; more of them usually means a bigger, more capable model).
Apple confirmed the partnership publicly. On the company’s earnings call, then-CEO Tim Cook said, “The personalized version of Siri is a collaboration with Google,” and added that Apple would “continue to run on the device and run in Private Cloud Compute, and maintain our industry-leading privacy standards in doing so,” as reported by 9to5Mac.
In one line: the voice and the app are still Apple’s, but the brain behind the hardest questions is now rented from Google.
Does Google see what you ask Siri?
Mostly, no. Most Siri requests never reach Google at all. Only the hardest ones do, and Apple says it strips out your name, Apple ID, contacts and location before the request leaves. Google is meant to receive a bare question with no person attached to it, use it once, and keep nothing. Apple also bars Google by contract from training future models on Siri data.
That is the design. The important detail is that “anonymous” here means Apple removed the identifying parts, and you are trusting Apple that it did so correctly.
Where does your Siri request actually go?
Apple showed a three-step system at its June 2026 developer conference. Every request you make gets sorted into one of three lanes:
- On your iPhone. Simple things — timers, reminders, opening apps, short text summaries — are handled by a small model running on the phone itself. Nothing leaves the device.
- Apple’s Private Cloud Compute. Medium-difficulty requests go to Apple’s own servers, which run on Apple-designed chips. Apple says these servers forget everything the moment your answer is sent.
- The Google Gemini model. Only the heaviest reasoning — long, multi-step questions — reaches the custom Gemini model, after your identity has been stripped out.
You do not choose the lane. The system decides, request by request, and it happens in a fraction of a second. That is one real limit on user control: you cannot tell Siri “keep this one local”.
What is confirmed, and what is only a promise?
This is where honest reporting matters, because these two things get mixed together.
Backed by technology you can inspect: Apple’s Private Cloud Compute is unusually open for Apple. The company publishes the software running on those servers so outside researchers can check it, keeps a tamper-proof public log of what code is deployed, and pays bug bounties to anyone who breaks its privacy claims. Apple describes all of this on its Private Cloud Compute security page. Craig Federighi, Apple’s software chief, put it plainly at the June event: “Data is only used to execute your request, and outside experts can continue to verify this promise.”
Backed by a contract only: Google’s agreement not to train on Siri data is a legal promise between two companies. There is no button on your iPhone that enforces it. The exact method Apple uses to scrub personal details from a query is also not open to outside checking. Reports disagree on one more point worth watching: some say the heaviest tier runs inside Apple-controlled infrastructure, others say those requests land on Google’s own cloud hardware. Apple has not published a full public breakdown.
The takeaway: the parts Apple runs are checkable, and the parts Google runs are a matter of trust.
Is this worse than using the Gemini app directly?
No — it is clearly better for privacy. When you type into the Google Gemini app with your own Google account, Google knows exactly who is asking. Your chats can be tied to that account, and human reviewers may read samples to improve the service. Siri is built to break that link before anything reaches Google.
So the fair comparison is not “Apple versus Google”. It is “a stripped-down question with no name on it” versus “a question tied to your account”. Siri is the first one.
What can you actually do about it?
You have fewer switches than you might expect, but you are not helpless.
- Check your Extensions setting. In Settings, under Apple Intelligence & Siri, there is an Extensions section that controls whether Siri can hand requests to outside chatbots like ChatGPT. Turn it off if you want fewer outside services involved.
- Turn off the summary features you do not use. Message and notification summaries can be switched off separately.
- Do not assume it is private just because it is voice. Treat a spoken request the way you would treat typing into a chatbot. If you would not paste a client’s medical file into ChatGPT, do not read it out to Siri either.
- Watch what your employer allows. Work-managed iPhones can have these features locked down by IT, and some companies will do exactly that.
If you want the wider picture on how everyday AI assistants quietly collect what you say, we covered a similar problem with meeting tools in Is Your AI Meeting Notetaker Leaking Your Calls?, and the account-level risks in Hackers Can Use Your AI Account Without Your Password.
Why this deal matters beyond your iPhone
Apple built its brand on the line that privacy is a human right. Now its flagship AI feature depends on the biggest advertising company in the world. Apple’s answer is to wrap Google’s model in Apple’s own plumbing, so the useful part arrives without the data trail.
If it works, it becomes a template. Other companies will license a strong model instead of building one, and sell the wrapper as the privacy product. If it leaks, or if the contract quietly loosens, it will be the clearest test yet of whether “private AI” means anything at all. John Ternus inherits that question in his first week in the job.
Frequently Asked Questions
Does Google get my name or Apple ID when I use Siri?
Apple says no. It removes your Apple ID, contacts, location and other identifying details before any request reaches the Google Gemini model. Google is meant to see only the question itself, with no person attached.
Do all my Siri requests go to Google?
No. Most requests are handled on your iPhone or on Apple’s own Private Cloud Compute servers. Only the hardest reasoning tasks are sent to the custom Gemini model, and the system picks the route automatically.
Can Google train its AI on my Siri conversations?
Apple’s agreement with Google forbids it. That protection is a contract between the two companies rather than something built into your phone, so it depends on Google honouring the deal.
Can I turn the Google part of Siri off?
Not directly. You cannot force an individual request to stay on your device. You can turn off the Extensions setting that sends requests to outside chatbots such as ChatGPT, and you can disable specific summary features.
When does the new Siri actually arrive?
Apple is expected to show the rebuilt Siri at its September 9, 2026 event, shipping with the new iPhones and stable iOS 27. Some regions, including the EU and China, are getting it later than others.
