Personal data harvesting for contextual ads and content should be a beautiful thing. They do it privately and securely, and it's all automated so that no human being actually learns anything about you.
Leaving the "privately and securely" bit to other commenters who will roundly correct you, I'm sure.
I've done personalized targeting of ads, and it is not necessarily a beautiful thing. The problem is a mismatch in the objectives of the advertiser, the objectives of the consumer, and the GDP maximizing outcome.
The GDP maximizing outcome is the thing that maximizes the total satisfaction of wants for the entire society. In theory, that should match the objective of the consumer. In practice, it does not, because the consumer is rarely perfectly informed or perfectly rational. Flaws at this level result in reduced consumer satisfaction, which result in reduced economic activity, and lackluster GDP growth in the long run.
In advertising, these flaws can be either explicitly or implicitly induced. The explicit way is the world of Edward Bernays and the world of PR; a fascinating subject in its own right. The advertisement targeting mismatch is about how success is measured and iterated into the targeting algorithm.
The personalized advertiser's objective is generally either to maximize revenue or earnings during the run of the ad campaign. This results in short-run oriented behavior which can be significantly mismatched with maximal satisfaction -- not necessarily intentionally, but because the system has no regard for satisfaction. Consumer satisfaction doesn't go into the algorithm explicitly and since campaign success can be most easily measured in the relative short run (did this impression result in a sale during the 30 day window that the customer is considered "owned" by this ad campaign), long run satisfaction cannot even show up implicitly. Most notably, impulse purchasing is strongly favored by the most profitable ad personalization strategies.
Ad personalization is good for short term revenue or earnings (or whatever is being measured), but it is not very good for long run GDP. From a strict economic standpoint, algorithmic targeting optimizes for flashy, shoddy products.
I know, because I did it.