Recommendation engines work on vast amounts of data they have on you and whatever made you speak about thing X was likely preceded by your internet activity which is not very unique as a precursor to speaking about X. In other words, if other people do Y on the internet and then end up doing stuff related to X, the recommendation engine will show you X just because you also did Y.

The other explanation is one of your contacts who were part of the conversation did things that either directly related to thing X, which you spoke about, or something the algorithm see other people do that relates to X, and you got shown ads based on your affiliation to this person.

I've also worked at FAANG and never seen proof to such claims anywhere in the code, and with the amount of people working there who care about these issues deeply I'd expect this to leak by now, if this happens but is siloed...

> I think it's true since far too many times obscure things I've spoken about appear in ads soon after the conversation

People have been making claims like this since at least the early 90s, about TV then, and no one ever credibly claims to have worked on something like this. I've worked with purchased ad data and I've never seen this data or anything that implies that it exists. It seems far more likely that its a trick of memory. You ignore most ads you see, but you remember ones that relate to odd topics that interest you.

I agree with this sentiment, people talk about x product then realise they are seeing ads for x product. Most likely the ads were there first and the people only start talking about it cause the ads have been working.

That’s pretty much it. You see an obscure ad without realizing it and have a related conversation later. Then when you see the ad again and make note of it, it feels strange.

Yeah, we're well past a point where "phones" have NPUs powerful enough to locally process "sensor" input and produce decontextualized probabilties of potential interests.

It's going to happen sooner or later and people will accept it, just like they accepted training of AI models on copyrighted works without permission, or SaaS, or AWS/PaaS, or sending all their photos to Apple/Google (for "backup").

I really question the commercial value of that kind of data. Credit card data has a lot more to do with intent to make future purchases than any keyword you might spit out verbally or in a search engine.

Just a personal anecdote: I don't have a dog, but my grandma has two. Once, while visiting her, the dogs were barking a lot. Almost immediately I started receiving ads for dog food in my cellphone.

It is more likely your GPS placed you in the vicinity (regularly?) with another AD ID that regularly searches for, purchases, or visits dog centric locations. It's also entirely possible that the other AD ID's (your grandma) dog food schedule is predictable and you happen to be visiting within a time frame of dog food purchases.

My grandma never owned a cellphone, only an old landline. And she buys dogfood in the neighborhood mom & pop store.

Well, we know for a fact it wasn't your mic being recorded. Maybe you walked by WiFi networks where people purchase dog food.

Or maybe the mic IS being recorded. We don't know it for a fact until all phone software is open sourced.

Reminds me of the chrome bug I filed years ago that is still unfixed. An extension with access to all browsing tabs can open a hidden iframe to a website that commonly would have mic and camera permission (like hangouts.google.com), and then inject its own JavaScript into that hidden iframe to capture mic or camera.

For this to work hangouts.google.com had to not include the HTTP header to block iframing but thankfully if you make up a URL the 404 page served on that domain does not include that http header.