Serious question though, why can't DDG do the same thing without compromising privacy? It would seem like it'd be possible to keep a log of things like "A user searched for 'movie about kidnapped daughter' and clicked on 'Taken'. Maybe weight that one higher next time" -- without keeping user details at all.
As an example: One really useful data source that I'm sure Google uses heavily is query reformulations. If a user does a search for Q, and then later does different queries Q' and Q'', and finally clicks on a search result, that's evidence that the result was actually relevant to the original query Q -- the search engine just wasn't smart enough to return it the first time around.
By itself, one data point like that is extremely weak evidence; the later queries might actually have been completely unrelated. But that's a source of error that tends to disappear when averaged across a large number of different users. In the aggregate, the data can be extremely valuable. But doing that kind of analysis requires correlating multiple searches from the same user, and storing the resulting profile for a long enough period of time to do useful aggregation.
You can't count on a browser session to be particularly short-lived. Even if you don't tie query logs directly to a username, the mere act of correlating different queries from a user inherently compromises privacy, as demonstrated by the AOL leak.
