The Problem with Personal Search

The hypothesis of personal search is that by looking at a users history, we can better predict, filter, and present search results. It is this thesis that has driven major search engines Google, Yahoo, and MSN to push the boundaries of privacy concerns to tailor their search results to each individual. I believe there are some inherent problems with Personal Search – not in it’s execution, but in the very premise upon which it is based, that people want personalized search results.

1. Consistency Breeds Trust:
   The most obvious shortcoming of personal search is that it delivers different search results to different people. As a search engine marketer, I have dealt for years with clients who are befuddled that different Google datacenters could cause their site to show as #1 for a keyword where they live, and #3 where I live and our rankings reports are generated.

   As personalized search has crept into SE algorithms, these problems have become far more persistent, of greater variance, and affect individual searchers, not marketers. Telling someone to “Google it” is no longer guaranteed to deliver a particular result, killing the brand identity. For a while, Pontiac advertised that you should simply “Google” pontiac, rather than visit their website. However, as personalized search becomes more pervasive, this type of free advertising for Google will fall by the wayside, as businesses cannot be certain of the result of such a search, even if their site is by far the most relevant and popular by Google’s standards.

2. Searches Are Often Not Personal:
   The singular act of personalizing results lowers relevancy, rather than raises it, when the user is searching for information on behalf of family members and friends. Eye opening research by the Pew Internet & American Life Project found that a online health searches are often on behalf of family members and friends, not for the searcher him/herself [2003]. Individuals using a search engine want THE best results, not MY best results, especially in situations where the ubiquity of the answers the engine serves up is essential (such as making a decision on a restaurant to attend).

   Example: Joe wants to take his visiting in-laws out to a nice Italian dinner. Usually on a budget, he has spent the last several months developing a search profile of finding cheap restaurants and visiting their sites via Google and Google Adwords results. When the day comes that he searches for “Best Italian Restaurant Durham NC”, instead of getting one of our best like Pop’s or Il Palio, he gets Macaroni Grill and Johnny Carinos, the restaurants he had visited in past. Joe could not trust the search engine to deliver an accurate result because the algo was based on him making a decision only relevant to himself, and not to group which he was trying to serve.

3. Results Should Remain Democratic, not Pander to the User:
   This, I believe, while nuanced, is the most problematic for personalized search. As a user’s search becomes more and more personalized, the plurality of the web is filtered. Right now, if I search for Social Security in Google, I find government sites, issue sites, encyclopedic results, etc. If I were a Senior Citizen, I may come to use the government pages more often, especially if I am at a time in my life when I first become reliant upon Social Security and need additional information.

   However, the temporal nature of my behavior is overlooked and, perhaps, the results become filtered and I lose the results that, while not necessary now, inform me that there is disagreement about Social Security, that there are Books and News that may be important at some time.

   The Democratic nature of search results, based on links, is replaced by an algorithmic approximation of your personal interests or characteristics. Instead of showing you the most relevant, popular results, they attempt to pander to the user when, often, the user is interested in a breadth of information well beyond what their demography would suggest.

Perhaps a compromise is in order: it certainly makes sense to use this information to alter advertising, or in whatever way improves click-through and conversion rates for AdWords customers, and perhaps a “personalized” tab could be available on Google, or inserted in the same way that books, news, products and images often are.

However, considering the rampant privacy concerns associated with personal search, and the shortcomings discussed above, we must reconsider the value of using demographic and historical user data to alter search results.

No tags for this post.