Amazon has one of the best on-site search capabilities we've ever seen. But surprisingly, the reason why it works so well is likely to be the same reason why Search *won't* work well on your site.
Want to read The Maltese Falcon? Type it into the search engine and Amazon instantly presents you with Dashiell Hammett's classic novel. Want to listen to Britney Spears? (After all, somebody must!) Type it in and you're looking at her latest albums. (Interestingly, when we just tried it, Amazon also suggested we purchase a set of headphones. What should we read into that?)
Based on just these types of searches, one might assume Amazon works almost 100% of the time. In fact, that's a comment we hear often when we talk about our research on how Search works. People are always telling us that it's not the fault of Search because, after all, Amazon works every time they try it.
Amazon's searches for books, CDs, DVDs, and videos work well because the content is what we call "Uniquely Identified." Users can easily search uniquely identified content because they know what they are looking for.
For example, people identify books by title and author. They identify CDs by artist, title, label, and songs. Almost every time we watched someone shop for a book or CD on Amazon, they typed in one of these identifiers.
For instance, when shopping for a book, one user typed in "Sum of All Fears" and Amazon returned seven different editions. Amazon didn't suggest any other books with "Sum" or "Fear" -- just seven editions of that one book.
When users search the uniquely identified content and those users know what those identifiers are, then Search works very precisely. In a study we conducted with 35 online shoppers, Search returned useful results 99% of the time for CD's and videos. However, for non-uniquely-identified content, such as toys, apparel, or pet supplies, Search only worked 31% of the time.
This is where Amazon starts to get into trouble. In addition to selling books, they also sell electronics. What did users type in when they were searching for a DVD player? Well, they didn't type in "Panasonic DVD-RV31K DVD Player (Black)," which is the actual product name. They didn't even try to type in "Panasonic," the manufacturer.
In our study, when users were shopping for DVD players, they typed in "DVD player." This is typical for non-uniquely-identified content.
When looking for a pair of Frye boots, one user typed in "boots." Another user, looking for colored pencils, typed in "craft supplies." Another user, looking for pearl earrings, typed in "jewelry." Not even "earrings," but the very generic "jewelry."
While there are non e-commerce sites that have uniquely-identified content, they are rare. The US Patent and Trademark Office, for example, allows users to look up trademarks by attributes such as name, holder of the trademark, and the attorney of record. (Search for "James Spool" under the attorney of record and you'll get a peek into the work of this author's father.)
The PTO is the exception, however, not the rule. It is more likely that the majority of content on your site will fall into the non-uniquely identified category.
Even within uniquely identified content, there are still exceptions when users want to find something by a means other than the identifiers provided. For example, we had a user who, having listened to Celtic music every morning on the radio, wanted to purchase a good introductory CD to the genre. Typing "celtic" into Amazon's Search revealed 889 results, providing no sense as to which one would be a good introduction.
Therefore, you can safely rely on your site's Search when you meet these three conditions:
(1) Your content is uniquely identified, and
(2) Your users are familiar with the identifiers, and
(3) Your users want to use those identifiers as the mechanism for locating your content.
However, if you don't meet any of these conditions, you'll need to find another way for them to succeed.
Not sure whether you meet these conditions? Look no further than your search logs. If you spot a lot of category names, like jewelry or men's pants, instead of specific content references, then you should take comfort that your content is like 90% of the content out there: non-uniquely identified.
