Ask HN: Is Google recording our conversations to serve us with relevant ads?

27 points by sarim ↗ HN
I believe we have all been served with ads for products and services that we haven’t even searched for. I presented this question to our Google Account Manager and he told me that Googles machine learning has become really good at predicting people’s intent. Having dabbled in Machine Learning myself, I don’t buy his crap.

46 comments

[ 0.24 ms ] story [ 99.6 ms ] thread
I'm sceptical, but it's hard to prove or disprove. So many other factors that go into making ad decisions (what did people who are somehow similar to you show interest in? Your friends? What's currently trending in certain areas of the internet, etc.)

I've only seen one attempt (shown in a video online, the topic was 'dog toys'), where recorded conversations were claimed to influence ads. Lots of people indicated foul play in the video, so I didn't book it in my "that's happening"-bucket.

I've only tried to reproduce it once, half-hearted, and failed.

The only problem with the parameters you listed is that you get served with the most random ads at the most optimum moments. I once saw an ad for Swiffer mop moments after mentioning casually to my friend that I am thinking of buying one.
That of course is suspicious. Do you sometime receive random ads that are completely out of the blue that you haven't talked about? (I'm asking because if you get shown 100 random ads per week, for how many can't you attribute them vs. how many seem to be related to some (online, offline) activity? Is it only by chance, and we just remember the 0.1% where it worked?)
I completely understand your point. Maybe I am noticing those ads only after I have consciously made the decision of buying the product. Or maybe the ads of that product are sub consciously nudging me to buy those products. Most of the display ads are meant to do just that
There's a cognitive bias toward noticing things that coincide like that, so it's quite possible you are getting Swiffer ads other times but notice it when you were just talking about it.
> you get served with the most random ads at the most optimum moments

If coincidences are just coincidences then why do they feel so contrived?

How many things have you casually mentioned and not seen an ad for? Maybe the Swiffer ad stood out to you because you had just talked about it? Maybe the reason you were thinking of buying one is because you saw an ad for it previously?

Don't think companies are willing to test if they can go that far. But many small apps are for sure
The simplest way to prove it is to show that kind of information flowing from a device to Google (or any other party). No one has ever done that. You wouldn't expect a full audio stream being sent, rather the device would look for triggers, but they would still need to be communicated in someway over the network.

Google (or any other company) have also never promoted such an ability to advertisers. If it would be a very closely guarded secret you would think that high budget advertisers would be the likeliest to have access to such targeting. However in "tests", like the 'dog toys' video that mikejb mentions and many more, the advertisers are usually no-name companies with cheap products. Most of these proclaimed tests also seem to be YouTube videos trying to get revenue from views. Furthermore, if devices would be looking for trigger words (like mentioned in the first paragraph), it would be even more likely that you would only see this happening for high volume/big budget advertisers.

I worked for a very large advertiser until end of 2017 and there was certainly no such offering.

Disclaimer: Today, I work for Google, but nowhere near ads. And I wouldn't share internals anyway.

> And I wouldn't share internals anyway.

No matter how disturbing they would be to the broader public?

What the broader public thinks is disturbing and what actually is unethical are two very different things.
> What the broader public thinks is disturbing and what actually is unethical are two very different things.

Maybe there are some things that may be unethical without being disturbing, I'm not sure the opposite is true; if the majority thinks it's disturbing then it's most likely unethical.

Facebook lets you target people for advertising based on the job titles Facebook users enter into their profiles.

So you can advertise Bingo Card Creator to teachers. Or swimming suits to lifeguards.

A few dozen people put gross things like "jew hater" as their job titles. So, if you trawled hard, you could target these "jew haters" on Facebook.

Some blogger screenshotted this and it became a national uproar about the evil Facebook profiting from hatred. None of the popular articles pointed out the nuance. And in response, FB nerfed the job title feature to hell, which means its harder to actually put relevant ads in front of people.

So that's an example of something that disturbed the public but was completely misunderstood.

> put relevant ads in front of people

I'd say the main problem is this. Nobody has ever asked for ads to begin with.

The unethical part is not better ad targeting, it's simply the fact that advertising exists and is depriving the world of millions of man-hours that could be put to better use.

People ask for varied content, but don't want to pay for many providers, so ads help provide the content.
Maybe people don't pay because they didn't want the "content" and it isn't good content anyway, but are merely being manipulated into watching it (through "feedback loops" like social media "likes", etc) by the creators of this garbage content for the purpose of getting ad views?
(comment deleted)
Wouldn't they just be linking keywords to you just like they already do from your searches? Therefore I don't think this would necessarily be a different product to promote, it's just another source of data which makes their targeting more accurate.
Anecdote time... So, I was sitting in a car with a colleague who was talking about how he used to live in the UK (we're both Danish but were speaking English because we had a non-Danish colleague in the car with us). My colleague talked about how he'd developed a fondness for cricket while living in the UK, to which I replied that I only really knew cricket from Hitch Hiker's Guide to the Galaxy and that it was one of those sports that I'd always been sort of curious about since it gets so little exposure here in Denmark. We agreed that we should maybe watch a game together so he could explain it to me and the conversation naturally changed topic shortly after. But a couple of days later Google Now showed me a notification saying that cricket results were available asking if I wanted to subscribe.

There's of course always a chance that I've done something that—unbeknownst to me—has led Google to believe I was interested in cricket, but I definitely didn't google "cricket", watch videos about it, mentioned it in emails or some such.

Google knows that you and your colleague met, drove together etc from location of both your devices. So, "maybe" your colleague frequently searched about cricket on his device. Based on these data points, google might have shown you cricket suggestions.
That's actually a really good and plausible explanation. Hadn't thought of that. My colleague is an Apple guy but I guess Google could still easily and legitimately know his location if he has Google Maps or other Google software installed.

But then I wonder why the 3rd guy in the car didn't get any similar suggestions.

I don’t know, but this sounds as bad to me as if they are actually listening. There are plenty of things I search for that I wouldn’t want to exposed to my friends or colleagues via this approach. I can think of ways to mitigate the violation of privacy, but that is what this is, if it exists.
Here's a live test that may help you to believe https://www.youtube.com/watch?v=zBnDWSvaQ1I and https://www.youtube.com/watch?v=SmM9ch_oXA4

In first video, a guy talks about dog toys and then he'll see ads related to what he talked about.

Worth noting, he doesn't have dogs or searched about it. The ads are even specific to the toy colors he talked about.

A more substantial test would be to monitor network traffic for audio streams, or if basic parsing is happening on device, then those keywords being set.
If it exists, then the parsing is almost certainly happening on-device, and the messages are encrypted and stenographically hidden.
This would have impact on battery life that no one has yet noticed.
Forget ads, are they selling it to Alphabet Agencies?
Everyone is worried that Google and/or Facebook are listening to us. Nobody seems to realise that it's actually worse if they _aren't_ listening to us, as it proves that their targeting is so good that they make you think otherwise.
A funny anecdote, when Google shows off the link to the "here's what we know about you page" every now and then, users more often then not will actually correct parts that are wrong.

You can find it in Account > Personal info & privacy > Ad Settings. You can also turn off ad personalization.

Yes, Google is willing to pretend to know only what you tell them to pretend they know. After Chrome made any login to a Google service also log you in on the browser, I decided to use a non-Google browser for any Google service. I switched over to another computer, which had a non-Google browser (FF) on it, made sure all cookies, etc., were deleted, and went to YouTube. YouTube said I wasn't logged in.

And yet, YouTube's recommendations were combination of obscure items of interest only to me that it had been recommending on the other computer. It was telling me to log in but already knew who I was.

Presumably it matched my IP address, but if I had used it for a while from a different IP address, it would have been able to match my usage and identified me that way. And if not that, one of my many Google-employee neighbors mentioned that they record your typing rhythms when they can, and those rhythms eventually uniquely identify you.

So when Google's "here's all we know about you" purports to reveal all, it shows you what they think of you rather than what they actually know about you.

Google's "what we know" is what they're confident enough in to bucket you to advertisers. Stuff like Youtube recommendations is a different algorithm.
IP is almost irrelevant, it’s your browser that is the major fingerprint. How many browsers will have your exact user agent, screen resolution, plugins installed, platform, timezone, language...? As soon as you log on from a specific browser, that browser is now associated to you whether you stay logged in or not.

Look at amiunique.org or panopticlick.eff.org for more details.

Thought police!!!!
I wish. I don’t get relevant ads from Google.

Even when searching for specific items “tacos” the ad will be stupid. I can’t recall the last time I used a Google ad productively.

Secretly recording audio is illegal in many (but not all) jurisdictions, including California. However, just because Google isn't recording you to target ads doesn't mean no one is.

A fair number of people are claiming they've been targeted with ads seemingly based on information that could only have been obtained through audio recording. But so far no one technically sophisticated has, and so these claims never seem to involve narrowing it down to a specific device, or checking the devices for malware or suspicious apps.

Try to buy an ad served against this data. How do you do it? If you can’t you can reasonably assume it’s not tracked.
Google doesn’t mention the source of data. You can setup display campaigns to optimize for more conversions, click through rates etc. It’ll serve the ad based on either audience or goals
Well maybe continuous subconscious exposure to an ad is causing me to discuss a product with my friends and I am acknowledging the presence of the ad only after having talked about it.
It remains unclear. There are reasons to believe they are (it'd really improve their ad targetting, and there are some anecdata to suggest they are), reasons to believe they aren't (they must have some ethics, and if they got caught, the result would be Bad), and reasons to believe they can't (constant audio recording would cause a suspicious drain on battery).
>I believe we have all been served with ads for products and services that we haven’t even searched for.

Sure, including ones I've never discussed; simple (not even requiring ML) statistical prediction of likely interest based on interests of people with similarities in search history or other things that Google overtly has in its tracking of profiles could well explain that; with ML applied well, that gets even better.

Covertly recording conversations for ads seems to be an unnecessary assumption to explain any effect I've seen or heard decribed, so while it's not impossible, I don't see any reason besides paranoia to believe it is true.

1. We've discussed this question with сolleagues, and one of them discussed with his wife bout with the gloves for his son, and on the next day he a received ads with bout with the gloves. Аnd it was not like that once. I do not claim that it was google, may be other ads company, but it is a fact.

2.Once a work colleague said me "Hey, listen my phone (nexus 5x)". From conversational speaker we heard as other people said! nexus 5x was on standby state! According to the conversation, I think that people did not talk on the phone, but offline.

sorry for my english ).

Did the wife Google the gloves, or open a page about the gloves that also had Google ads (or Analytics maybe)? Ad companies do know that certain groups of users/devices share interests.
There's a few data-grabbing questions I've always wondering with Google and Facebook and others.

Does the Facebook app, if granted permission to access photos, upload thumbnails or metadata generated on-device that's descriptive-enough to characterize photos located on devices even if they're not chosen by the user to be uploaded to their service.

Does Facebook and/or Google send your current clipboard contents to their server? Google Maps seems to do-so since it has the address pre-filled when you launch the app if you've copied it from elsewhere.

Anybody able to officially answer or speculate on either of these? I've always resorted to assuming they do because they can.