Last week, Udi Manber, Google’s VP of Engineering, gave a short presentation where he discussed the difficulties that face search engines.
One of the problems he pointed out was that “20 to 25% of the queries we see today, we have never seen before.”
A patent application published for Yahoo on the same day, Interactive search engine, about supplying predictive queries to searchers as they are typing the query terms into the search box, bases a decision to do so on the assumption that
…it is highly probable that a user intends to issue a query in which at least one other person has issued previously.
I’ve written about Yahoo’s approach to predictive queries before, as well as Google’s use of predictive queries.
Both methods rely upon looking at the popularity of queries used previously, as well are frequent recent queries. Yet if so many queries are new, does the assumption becomes somewhat questionable?
The presentation of predictive queries on mobile devices may mean less typing with a stylus or numerical keyboard, and may mean that people performing searches on laptop and desktop computers have a little less typing to do.
There’s probably some value to continuing to use predictive queries even if so many searches are new and unique, and a percentage of searches aren’t answered by predictive results even if they aren’t unique.
The new patent application describes some potential uses for predictive queries that we may not have seen in action, such as their use with different types of searches, perhaps in conjunction with tabs, such as image, video, or shopping searches.
Some other suggested queries may be shown because they are “related” to predictive queries, but without being completions of a spelling of the word being typed into the search box.
If a quarter of searches are new, then predictive queries aren’t going to help those searchers. But they could still be useful to other searchers.
Predictive queries could still help some of those unique searchers. There are many ways to say the same thing and while you might type a query one way and I another either one of us might happily select the other’s query if it presented itself to us.
I’m thinking of two long queries that maybe differ by a word or two. Would the queries be different if you typed “an” and I typed “the” and would you get fundamentally different results for each.
Instead of typing in a long query might you not just select the nearest match that was presented to you.
I agree predictive queries won’t be useful to everyone, but I wonder how unique all those unique searches really are.
We were discussing this a few weeks ago at Seomoz.
identity came up with a brilliant solution. He proposed the use of patterns to predict invisible long tail queries.
I think it is perfectly possible and useful to predict head and long tail search queries.
Right now, these predictive queries seem to focus primarily upon what a searcher may have intended to type, rather than what they may be searching for.
I suspect that we may start seeing some different approaches to query refinements showing up in the dropdowns that these are presented in.
Pattern matching is discussed in the patent application, for such things as part numbers – it may be useful for other areas.
Unfortunately, the patent office web site is returning an error message instead of the actual patent application. I hope it’s a temporary glitch.
I was involved with this project too. I never thought I would be, but for my study I had to be there.
I never thought I would be, but for my study I had to be there.
Hi Bill,
in my opinion this topic is highly interesting simply because it leads into the ying and yang of SEO.
SEO consultants were used to use the Adwords keyword tool in order to determine more relevant keywords. Predictive queries as they are presented by Google Suggest (http://www.google.com/webhp?complete=1) or in the Firefox search form are the next best things to take a closer SEO look on.
Why? Simply because the query/keyword data are completely different compared to the one of the Adwords tool and simply because on can see at once how many web page are indexed for the different relevant queries/keywords.
Google predictive queries do not only show similar queries or complete them in a senseful way. They preferably show queries with high result numbers as if this would be a criteria for relevancy …
What does that mean in the long run? It means that Google predictive queries have a tendency to be a self fulling prophecy.
Instead of starting queries for what they are really looking for people will more and more use predictive queries which will lead to web pages especially optimized for them.
We should save the last 25% from being predicted.
Hi Dirk,
I share your concern that the suggestions shown may unduly influence people to settle for looking at pages that are suggested rather than come up with their own queries to find the information that they are seeking.
I would love to see some data on how many people select one of the predicted queries, as opposed to choosing their own terms…