By "spread across IPs", I meant "from different physical computers", rather than the sorts of dynamic IP behaviour from a single host that can, as you say, be overcome through the use of cookies for tracking.

And how does Google know what kind of car you have? It knows what kind of car you might have been looking for information on, but that could have been because you were looking to buy one (and decided not to), or because you were helping a friend change the spark plugs. The data is considerably less rich when it's just search data without any ability to track conversions.

That said, I didn't actually think about Google Analytics when I originally replied. That is likely to be a source of data that is as good as Amazon's, if not better. Search-derived data on its own is of limited value.

It's not about a single piece of information. It takes one login before a new PC's history is linked to every other PC you have used and logged in from. Ok, linking to whichever account you use the most from that PC.

In the same way, Searching for an "Acura TSX pricing" in 07 says little. But, if you then search for "07 Acura TSX Owner’s Manual" in 09 you probably bought that car. I don't think Google algorithms are smart enough to say "bugatti veyron videos" relates to a desire for a faster car not the ownership of said car, but who knows. Google has focused on mining data for a long time and are fairly good at it by now. Search over time can say a lot, search for diapers in 00 and “toy’s for 5 year olds”, and you might have a kid born around 00 or are at least interested in buying stuff a child that age.

A lot of this is probabilities, thinking there is an 80% chance I have a 07 Acura TSX is a lot better than guessing I probably have a car. AKA showing me advertizing for a low end BMW is probably not a waste relative to the average user.

I usually don't login to google, hence my original assertion.

I've not seen any evidence that Google actually do any analysis that is quite that smart. My best guess is that they just do keyword analysis. This is something of an educated guess, because I know someone who worked on keyword extraction at Google.

Even if Google _could_ extract that kind of rich data you describe, I very much doubt they have done it to the point where it is in a format that is suitable for providing targeted ads (beyond what they already do, which as I said, I would guess is keyword-based).

The difference with something like Amazon's data or Google Analytics is that it is structured data that is already perfect for providing recommendations and targeted ads.