When you search google you are presented with two kinds of links.  Most of the links come from google’s webcrawlers and they are presented in an order that reflects google’s PageRank algorithm’s assessment of their likely relevance.  Then there are the sponsored links.  These are highlighted at the top of the main listing and also lined up on the right side of the page.

Sponsored links are paid advertisements.  They are sold using an auction that determines which advertisers will have their links displayed and in what order.  While the broad rules behind this auction are public, google handicaps the auction by adjusting bids submitted by advertisers according to what google calls Quality Score.  (Yahoo does something similar.)

If your experience with sponsored links is similar to mine you might start to wonder whether Quality Score actually has the effect of favoring lower quality links.  Renato Gomes, in his job market paper explains why this indeed might be a feature of the optimal keyword auction.

The idea is based on the well-known principle of handicaps for weak bidders in auctions.  Let’s say google is auctioning links for the keyword “books” and the bidders are Amazon.com plus a bunch of fringe sites.  If Amazon is willing to bid a lot for the ad but the others are willing to bid just a little, an auction with a level playing-field would allow Amazon to win at a low price.  In these cases google can raise its auction revenues by giving a handicap to the little guys.  Effectively google subsidizes their bids making them stronger competitors and thereby forcing Amazon to bid higher.

Of course its rare that the stronger bidder is so easy to identify and anyway the whole auction is run instantaneously by software.  So how would google implement this idea in practice?  Google collects data on how often users click through the (non-sponsored) links it provides to searchers.  This gives google very good information about how much each web site benefits from link-generated traffic.  That’s a pretty good, albeit imperfect, measure of an advertiser’s willingness to pay for sponsored links.  And that’s all google would need to distinguish the strong bidders from the weak bidders in a keyword auction.

And when you put that all together you see that the weak guys will be exactly those websites that few people click through to.  The useless links.  The revenue-maximizing sponsored link auction favors the useless links and as a consequence they win the auction far more frequently than they would if the playing-field were level.

(To be perfectly clear, nobody outside of google knows exactly how Quality Score is actually calculated, so nobody knows for sure if google is intentionally doing this.  The analysis just shows that these handicaps are a key part of a profit-maximizing auction.)

Renato’s job market paper derives a number of other interesting properties of an optimal auction in a two-sided platform.  (Web search is a two-sided platform because the two sides of the market, users and advertisers, communicate through google’s platform.)  For example, his theory explains why advertisers pay to advertise but users don’t pay to search.  Indeed google subsidizes users by giving them all kinds of free stuff in order to thicken the market and extract more revenues from advertisers.  On the other hand, dating sites, and some job-matching sites charge both sides of the market and Renato derives the conditions that determine which of these pricing structures is optimal.