Do you mean that advertising won't generate the maximum amount of revenue possible unless it violates users' privacy?

Correct. The crux of the issue in the economics of the web is that users privacy value is not explicitly part of the advertiser-publisher transaction. Advertisers want more data because it works. Publishers are ok with that as long as it doesn't bother their users too much. But it's not explicitly priced in the transaction, and users have no say in the transaction at all.

I cannot verify your models.

But I can verify that I think Google gets my interests wrong more than 9 out of 10 times to the point of repeatingly insulting me and my family, and they are supposed to be very smart so I guess they to have equally smart models.

If they had just stuck to something relevant to the pages I visit they'd probably hit more than 50% of the time. If I visit a video about fixing stuff I might be interested in tools, paint or parts. If I visit a local news source around lunch I'd probably enjoy ideas for dinner from local shops.

That said I probably wouldn't buy much more but at least they wouldn't insult me and my family.

I can throw around anecdotes as well. I can verify that I think the majority of ads on content websites that I see and find interesting often have nothing to do with the article that I'm reading.

But in the absence of data, I'm leaning towards trusting that a market-leading company incentivized and dedicated to optimizing targeting models has a stronger research on targeting the majority of users than vague internet claims of bad targeting. But what do I know.

Really? To me it seems that "user-predictive" modeling just mean "show ads for sites user visited recently". E.g. if I buy a blender on Amazon (and maybe check a few comparison sites or something before that) I'm going to get blender-related ads. If that's the best advertisers can do, it's pretty weak sauce... Why would I need 7 blenders?

It seems to have worked better than well for New York Times: https://digiday.com/media/gumgumtest-new-york-times-gdpr-cut...

Do you think it's not repeatable?

> except for privacy-minded users

...that's like saying "high standards of medicine are bad except for hygiene-minded users." The fact that the user has other concerns in their life doesn't change the fact that it's still bad for them.

It might be good for supply (publishers) short term, but it's worse for demand (advertisers) and for ads quality since it removes personalization opportunities. An open-market like RTB where offer meets demand is more efficient (except for privacy-minded users which aren't even part of the transaction). It's also worse for supply long-term as it ties them more to Google and FB, or to a non-scalable contract negotiation process mad-men style with a handful of advertisers.