[zmmmmm]

I spent 10 mins trying to find a clear statement of whether Google uses information submitted to Gemini for training and I couldn't find one. It is hard not to come to the conclusion they actively try to obfuscate it because there are many statements that vaguely sound like they should address it but then don't properly do that.

So I would have to suggest, use these features with extreme caution on any page you consider private if you aren't prepared for your private information to get sucked into Google's Gemini training data.

[walkingthisquai]

It's right there in the privacy center. The answer is quite unambiguously yes by the way. (This is for the Gemini app):

"How your data is used Google uses this data, as described in our Privacy Policy, to:

Provide our services Maintain and improve our services Develop new services Personalise our services (learn more) Customise our services Communicate with you Measure performance Protect Google, our users and the public

These uses extend to the generative AI models and other machine-learning technologies powering our services."

[simonw]

I find that answer ambiguous.

Does "Maintain and improve our services" mean "any private web page you ask Gemini about will be dumped into our training data"?

I've still not seen a solid answer to that from any of the AI labs that use language of that nature.

[afro88]

The wording is intentionally broader than just training. It encompasses training and anything else they want to do with your data in the name of "maintain and improve our services".

Safe to assume they will use your data for pretty much anything they can, including model training.

[walkingthisquai]

When you go to the Gemini in Chrome section on the same privacy page it states:

'When you use the Gemini in Chrome feature, Gemini collects and processes page content and the URL from the browser tab you’re viewing by default. Some of the page content Gemini uses might not be visible to you."

I think they're being reasonably explicit about what they're doing.

Note: I'm in the EU so not sure if this is what's shown everywhere else.

[gapan]

I don't think there is anything ambiguous about it. How else are they going to "personalize" their services, if they don't consume your personal data?

[simonw]

If I am looking at a web page that shows me my own API keys for a service in plain text, and I accidentally click the Gemini button while viewing that page, is there a chance that someone in six months time might ask Gemini for an API key for that service and have mine returned to them?

I'd love to get a confident answer to that question.

[reciprocity]

The parent comment has a point. A layperson (or at least many people not read into this topic) isn't going to read the language in that privacy policy and come to the conclusion that "any private web page you ask Gemini about will be dumped into our training data". The text on Google's privacy page could absolutely be made more explicit.

[rotis]

I'm sceptical a layperson will understand or care what it means that their data will be used in training. If you are concerned about such things this heavily implies you don't want to share your data. Just don't agree to the terms and move on.

[SirFatty]

Why not assume yes?

[hofo]

lol any answer to “do we use your data” that isn’t “no” is a yes

[wodenokoto]

We had legal trying to figure out if data submitted to Google Cloud was shared with Google, and the conclusion was that it is unclear from their TOS. Their TOS is a bunch of circular references to different agreements.

[freakynit]

I think it goes without saying that anything Google provides for free, they do it to garner user data. Traditional search is dying. And so is the advertising that comes with it. They are finding alternatives. They'll keep "injecting" themselves into everything we use regularly. Ads will get even more targeted... much more contextual in realtime.

[rkagerer]

I simply don't use Gemini.

I never asked for it, or agreed to anything having to do with it. I'm pissed off to find it bound to a hotkey or gesture on my phone (I'm still not clear what the actual gesture is that keeps invoking the damn thing).

The more unsolicited crap Google jams down the pipe at me, the sooner they're going to discover I'm not at the other end and the pipe feeds straight into a septic field.

If anyone from Google is reading this, I hope you're ashamed of your dark patterns. You used to be a testament to the ideal of putting the user first. Now I can't distinguish you from any other crummy, misleading, self-serving tech gorilla.

[rollcat]

This is the problem with being the biggest in X. Facebook desperately tried to branch out: phones, video, VR... Eventually the only thing that worked was buying other social media companies.

Google is in the same position, yes they have Android, GCP, Gmail/work suite, etc but even all of that combined couldn't sustain the moloch.

[freakynit]

Exactly.

These insane levels of revenue streams cannot be sustained with just one product for longer periods of time. Eventually, every one of these biggest X'es will need to branch out to different industries/domains to sustain those levels.

[jon-wood]

If Google had simply stuck at what they used to be, a solid search engine with some unobtrusive ads plus some other ancillary services, they could have sustained that more or less indefinitely. What can't be sustained is the endless growth that the stock market demands. If you dare to say "you know what, this is probably enough" you'll be immediately punished by the market.

[johannes1234321]

We are seeing the counter proof right now. Search in itself isn't sustainable. The concept is challenged by GenAI-based approaches.

Of course they are not a 1:1 replacement, but for the first time we see Google's model being challenged and them having to defend, which they do by trying to drive competition of by integrating genai into their products. Once the competition is gone, they can reconsider.

[2muchcoffeeman]

There’s plenty of services that they could have gone into and charged money for.

[lwhi]

I think this is true for all of the big orgs.

Main difference is that Google is bad at monitising any (non advertising) service; so free becomes the main proposition.

Is Meta much different though?

[jsnell]

Google's non-advertising revenue is about $25B/quarter. There's only a handful of tech companies with higher revenue than that.

It is for example a bit more than Tesla, double that of Broadcom or Oracle, a bit less than Dell or TSMC.

Seems they're actually pretty good at monetizing non-ads goods and services at this point.

[lwhi]

A good point, but I still think they're struggling to innovate.

Their advertising revenue is over $71B/quarter.

As far as I know the majority of the non advertising revenue is from Google Cloud and subscriptions for services.

Most of the interesting product initiatives they've launched (that I can think of) have folded.

[noduerme]

So, naive question: If you click this button while looking at your bank account or, say, a mortgage application form, or a government website where you're paying taxes, etc... is all your form input literally just sucked into some insecure dataset in the cloud used for training Gemini?

[zmmmmm]

I would assume they keep it secure as far as the training data goes, I don't have too many doubts about that.

But aside from that part, yes, I can't see how you could make a different assumption.

[kmod]

They are definitely capable of writing such statements, which you can see in their enterprise products. In my Google Workspace gemini app it says pretty prominently and clearly:

  Your [ORGNAME] chats aren’t used to improve our models

The Google Workspace privacy hub is similarly easy to read and clear that they don't train on your data: https://support.google.com/a/answer/15706919

So they definitely understand that people want to hear that their data isn't being used for training, and they know how to say it clearly and reassuringly. Which makes the omission of that in their consumer products more telling in my view.

[doctorpangloss]

Haha, but do they paraphrase your chats, and use it for training? (Ye$)

[j_timberlake]

I think if Google trained current models on private data, confidential info would leak constantly, it would be an absolute trainwreck. If Gemini leaked your Gmail and Chrome activity, Google would get sued and regulated into oblivion.

But Google needs to leave this option open in the future, in case they have to go all-in on an arms-race against China, if Chinese AI starts becoming an actual threat somehow. And it's easy to predict the USA gov would prioritize that race over privacy concerns.

[tgv]

For what it's worth, these are the current conditions under which it will be active:

* Be 18 or over and in the US.

* Use a Mac or Windows computer.

* Use the latest version of Chrome.

* Have Chrome’s language set to English (United States).

* Sign(ed) in to Chrome.

[rollcat]

It's a test run. They will relax those conditions first chance they get.

[acters]

As far as I can tell, Linux will remain not targeted by attempts to sponge off all kinds of user data. Which makes me so happy that I finally made the leap.

[IX-103]

From what I heard it's only not on Linux yet because they've had some serious crashes due to incompatibility with Wayland. Don't worry, it'll come to Linux in time.

[sieep]

firefox + DDG + linux + vpn is my preferred combo, or the 'privacy stack' as I like to call it.

[troyvit]

Thank you, thank you Wayland for being you.

[xnx]

> * Use a Mac or Windows computer

I wonder if Chromebooks fall under Linux

[jkaplowitz]

They definitely don’t fall under Mac or Windows, at least.

[rkagerer]

Could you please add:

* Opted in

(Or at least, "* Hasn't opted out")

[FL410]

This is literally the worst part of Gemini. I spent a lot of time trying to figure out if, and if so what, they are training on even with my stupid $250/mo subscription. It's totally opaque.

[holoduke]

Or course they are using it. With Google even this keyboard stroke on Android is used for something.

[mig1]

Ex-Googler here, at least in the UK, privacy was taken very seriously by all employees, we never collected data without explicit consent and never used it for anything but what the user granted permissions for.

[repeekad]

Also ex, for a reason, everyone with ethics left, and Google “is a conventional company” now

Edit: reference https://abc.xyz/investor/founders-letters/ipo-letter/#:~:tex...

[LunaSea]

Didn't Google release many data collection features as an "opt-out" setting (ie: without user consent)?

[troupo]

> privacy was taken very seriously by all employees, we never collected data without explicit consent and never used it for anything but what the user granted permissions for.

Except when you literally trick people into providing their data: https://x.com/dmitriid/status/1664682689591377923 by pretending that your dark patterns are "explicit consent"

Or except when you literally trick people into providing their data by connecting totally unrelated services https://x.com/dmitriid/status/1908951546869498085 and you don't stop until the person gives up and submits to you.

Except when your own support explicitly says that your behaviour is tracked across completely unrelated Google services: https://x.com/TeamYouTube/status/1849952594992435493

Except when you literally sign people into user accounts automatically with most data collection options turned on.

Except...

[frays]

This claim, without a source, runs the risk of being misinformation, which is a massive problem in 2025.

Can you provide a reliable source to verify it?

[interloxia]

Federated Learning of Gboard Language Models with Differential Privacy

It's not nothing, but it's something. And, at least on my phone, it's not obvious if it can be turned off.

https://scholar.google.com/scholar?lr&ie=UTF-8&oe=UTF-8&q=Fe...

[drilbo]

You can always opt out of using gboard altogether.

FUTO Keyboard is quite nice.

[interloxia]

Thanks for the suggestion. It supports multi-lingual typing which in a requirement for me. I haven't checked other keyboards for a long time so perhaps that has become more common.

The integration with whisper is nice too.

[kelvinjps]

I don't find the multilingual features as polished as Gboard, this is what prevents me from switching, in Gboard you can install multiple languages and write without having to switch and it will provide autosuggestions and spelling support based on the language you're typing without having to manually change the language

[troyvit]

Heliboard is another good one: https://github.com/Helium314/HeliBoard

[Physkal]

Any recommendations on android keyboards?

[drilbo]

I recommended FUTO keyboard in sibling comment. FlorisBoard is a nice FOSS option, but some features are still WIP. Personally, I've switched fully to ThumbKey, but that's got quite a learning curve.

[colonelxc]

https://research.google/pubs/federated-learning-for-mobile-k...

[dmesg]

Thanks for getting ahead of me. I add their competitor MS doing the same even more openly:

https://support.microsoft.com/en-us/topic/microsoft-swiftkey...

Always assume companies will gather, use and share your data in all ways they legally can. The burden of proof is never on the user that companies don't milk us. Calling it "misinformation" as someone further above did is bizarre. This is the default business model of big tech.

[48terry]

Phew, thank goodness someone policed this HN comment's quip about a multi-billion dollar company. May I recommend a more ambitious target in your war on misinformation next?

[SquareWheel]

Whether it's aimed at large companies or not, I'd still rather not see misinformation spread. People already have poor enough understandings of what companies actually do and don't collect. There exists ongoing conspiracy theories that phones actively listen to conversations while in your pocket, despite there being no evidence to such a claim.

Facts do matter, and I appreciate those that make an effort to state them correctly.

[rpdillon]

Yeah, the facts were stated correctly. They just didn't provide a source. That doesn't make it misinformation. It means it's a claim without a source.

But to answer the question, Gboard absolutely uses your data. And it's right there in its privacy policy.

[lwhi]

Misinformation isn't more or less appropriate depending on the target.

[redml]

being privacy centric is a badge of honor these days, so if they aren't making it clear or not giving an easy to find option, then it's a guaranteed to your queries and outputs are used for training.

[jacooper]

They do, even when you pay for gemini pro/ultra.

[blauditore]

Google generally uses data from free users to train ML models etc., but no data from paying customers. I don't have a link to back it up though.

[kirito1337]

they obvuscate it with legal nonsense