intermediaries that will be compromised
Except that YouTube doesn't need any of that to recommend Minecraft videos to you. One mundane explanation that seems more likely is that it's the type of content that on average works best on people they don't yet have information on.
Reminds me a little of these "phones listen to everything we say, otherwise I wouldn't have been shown this ad" anecdotes that don't hold up against empiric evidence.
Microsoft would definitely sell data on which IPs have Minecraft users, as would your DNS provider, cloudflare, or your ISP.
It could also be coincidence, though my experience is YouTube's suggestion algo is very tightly tracked to use data. (What I was fed on a guest account recently was a mix of fascist propaganda and AI nonsense masquerading as reportage.
If phones weren't listening technology like audio beacons couldn't have been invented or useful (https://medium.com/@williamwais01/ultrasonic-beacons-the-sil...).
Surveillance capitalism has been so successful because it's so opaque. It would take a whistleblower for you to know when and how the data companies have is used against you, or how they got that data in the first place. Their tactics can be used in ways that are highly targeted and transient, especially for data collection companies like Google.
Even for something as basic as search results what I see when I search may not be what you see, and what I today might not be what I see when I take the same actions tomorrow.
This can make identifying what our phones are doing almost impossible. A possible explanation for the "phones listen to everything we say, otherwise I wouldn't have been shown this ad" phenomenon might be that a phone picked up an audio beacon being broadcast while something played on a TV, which sparked a related conversation in the person carrying the phone. The conversation wasn't recorded, but topic being discussed was successfully logged anyway. There are countless other data points available that could be used in the same way. In cases like that it would clearly not be coincidence that Google showed an ad when it did, but the spying involved was even more involved and invasive than just listening to what was being said.
There's nothing to stop Google from having small clusters of phones listening to everything for certain periods of time under certain circumstances. They wouldn't have to send all that audio data back to their servers to be effective, just monitor for a sample of specific words/phrases (processed on device) and send back a flag when something is overheard. That kind of behavior would be extremely hard for researchers to catch.
The truth is that we're not allowed to know how and when we're being surveilled, but we are being watched all the time, and that data is collected to be sold or used against us at every opportunity. It doesn't do any good to tell the person imprisoned in the panopticon that he's being paranoid and that it's all coincidental. Even when it happens to be, feeling watched is the natural response.
Honestly I wish I would get more targeted ads for the stuff I look for, when I look. Instead, for some, it happens that after I buy them, I get repeated ads after couple of weeks or even months (like, invoice is on gmail, photos of the object on my phone, but nooo they want to trick me, so they still send me more ads of the same shit that I will not buy again in years).
Seriously, I think the tracking is as crappy as most software is. Sure, it might identify one/two keywords and throw ads at you, but it does it a dumb volume way that corporations work, not in a smart "we know everything about you way", that a true geek might implement.
Right now the amount of data about you companies have is just massive. Everywhere you are at every moment of the day, who you were with, what you talk about, everything you buy, and every website you visit, what your mood is, what your level of education is, how you react to stress, it's all too much information for companies to extract useful data out of right now. AI is going to help with that. Unfortunately, like everything else AI does, it will do it pretty badly and with lots of errors and hallucinations. The companies using AI won't care though as long it works enough times on enough people that they make money.
To make matters worse, ads are only a small part of what all that data is used for. It's the thing that's most visible to you though, so you can imagine that if your ads show that companies think the wrong things about you that HR departments and other companies you interact with offline probably do too.