The business of web search

I've been thinking a lot recently about the business of web search and where the up and coming semantic search offerings fit in. Google, Yahoo! et. al. make their money via the paid links and adverts that they provide to supplement the main search results. From their perspective not all user searches are equal; the difference is in the users intentions. Take the following list of the top 10 searches on ask:

1. MySpace 6. Online Dictionary
2. YouTube 7. Craigslist
3. Facebook 8. Cars
4. Pamela Anderson 9. Wikipedia
5. Angelina Jolie 10. Hannah Montana

Six of the top ten are navigational searches - the user wants to go to a particular website and instead of typing it into the address bar and adding a '.com', they type it into their search box and hit the top link. I'm guessing that "Pamela Anderson", "Angelia Jolie" and "Hannah Montana" are all examples of informational searches, the user is broadly looking for information about the particular search term, whether its recent news articles, biographies or images. Only "cars" looks like it might be the kind of search that could actually be monitised by targeted adverts. Take a look at the results for these in Google, a search for myspace provides no adverts - it makes them no money at all. Google generally does not allow you to advertise against someone else's trademark terms so naviagational queries are useless for them. Compare this to the search for cars which provides eight ads for some modest googly profit.

To make money in web search then, you need two things: first, a large inventory of adverts that can be accurately matched to searches where the user is at least slightly interested in clicking on an advert; secondly you need the users to actually come to your site (or use your toolbar etc.). Internet users tend to stick with just one search engine for the vast majority of their queries and it is because of this that navigational and informational queries are important - they increase the stickiness of the search provider. But this is only a second-order effect, the informational queries are not directly profitable, in fact they are more like adverts for the search engine itself.

Strangely though, it is these kinds of queries that semantic search seems to have latched on to as the main target. Take the top ten queries from powersets example questions:

who discovered the neutrino when did Caesar cross the Rubicon
what are the ingredients in beer what was banned by the FDA
how many people were evacuated from New Orleans during hurricane Katrina who plays for the Oakland Raiders
paintings by Salvador Dali what causes diabetes
actors in Forrest Gump movies starring Tom Hanks

Out of these, only "movies starring Tom Hanks" looks like it could possibly be a product search, the others are never going to provide any useful advertising click-through.

So why focus on these informational queries when the real money is tied up in the product queries? Perhaps its a solution in search of a problem - everyone knows that search is hot, so is natural language processing and semantic knowledge so why not throw them all together and see if Microsoft will buy you for $100million.