http://www.channelregister.co.uk/2008/03/18/when_google_does_evil/

(6 Pages - see links at bottom to progress)

This article discusses many of the details of Google’s ad auction service that do not get covered in Econ 204 lecture. Recently, advertisements on Google have been receiving less clicks, which Google claims is an attempt to get rid of “unintentional” clicks and provide higher quality ads. The article uses this as a jumping point for examining how Google’s secret ad machine works. In our discussions in class, we talked about how bids for ads might work for specific keyword searches. However, Google also uses something called “automatic matching” which tries to apply ads to related searches, even when the advertiser has not bid on that search term. That way, if the terms the advertiser bid on do not generate as many clicks as Google expected, they can broaden the appearance of ads to other searches and charge the advertiser for the clicks they receive with these other keywords. Google is currently treating this service as a beta and automatically applying it to advertiser bids unless they opt out, meaning many advertisers may be paying for ads they hadn’t bidded on.

Google’s algorithm for running the advertising auctions is much more intricate than the algorithms given in class. In fact, only the company itself knows exactly how it works, and it changes over time. One added intricacy is that instead of bidding on an exact search, like was discussed so far in class, advertisers can also bid on a “phrase match” (which consists of the exact search plus a few other words tagged on) or a “broad match.” With “broad match,” Google will place the ads in searches it deems relevant to the one the advertiser paid for. However, the article gives the example that a bid on “dog gifts” may result in an ad on the search “hot dog buns” which likely is not what the advertiser wants. Previously broad match meant applying the ads to searches with any keyword in any order, but now ads might appear on searches containing none of the keywords supplied by the advertiser. Erick Herring of Adapt is quoted in the article saying “If you’re Google, the way to expand revenue is to expand ‘coverage.’ Google can’t get paid if someone does a search and there’s not an ad in place that a person can click on.”

By using some hidden algorithm to determine relevant searches, Google may not perform the service the advertisers thought they was paying for, and companies may be charged for ads thay didn’t want to place. One customer for instance found out that after paying nearly $7500 a month, most of his ads appeared on other “content sites” that pay Google for their ad service rather than Google itself, some of which with nonsensical URLs like 1000jogos.com. The matching algorithm for matching advertisers to slots has some hidden components as well. In class, we treated this as a second price auction with slots being valued based on click-through rate and advertisers bidding on the value of a single click. However, Google also considers other factors like the quality of the page placing the ad and the relevance of ad text. They give each advertiser a quality score and the lower one’s score, the higher the minimum bid to place any ads at all and often the lower the placement on the page. Hence big businesses get better treatment than small ones and can place ads with a lower minimum bid. As a business, Google’s algorithms moreso act to maximize the company’s own revenue, rather than the social welfare maximizing algorithms in class.