Google Tries Behavioral Advertising

The buzz on the blogosphere seems to be that Google are now starting to track your keywords across multiple search queries, and serve ads accordingly. This type of behavioral tracking and ad-serving has been in use over at Yahoo for a while, although it”ll be interesting to see how Google do it.

The idea is essentially a very simple but powerful one; if I search for “Flights to New York” and then ten minutes later do a search for “Hotels”, Google will remember from my previous search that I”m going to New York… and serve ads that would match “Hotels… in New York”.

This news made me think about how we could implement this sort of behavioral tracking and ad-serving for video. Can we assume that because a particular user has watched lots of videos in a certain category, or with certain keyword tags, that he/she will follow the typical ad-clicking behavior of users with similar viewing patters?

Take for an example a user who we can see usually watches videos about cars, and has a higher than average click-through-rate when offered car-related commercials. Should we therefore serve that user car commercials even when he decides to watch a video about dogs?

What if we find that users who click on ads for alcohol tend to do so most often when watching videos about food - even though most other food-video viewers prefer ads about restaurants. Should we skip the restaurant ads for this user, and just serve him alcohol ads? Would this not result in a self-fulfilling prophecy; the user will click proportionally more alcohol ads because thats what we are offering him? And when tracking the response data, how does one define the boundary between cause and mere correlation?

But this is just the beginning of the possibilities; how about tracking which ads get clicked most on specific videos or when appearing on specific websites, and giving these greater prominence over their competitors on that particular content, even if on average across the network these ads are performing worse?

What if all this tracking and actually has no real impact at all?

My conclusion? Must investigate further.