Google Trends: Peer Into Google's Database Of Searches

Now live via Google Labs is a new Google Trends service, announced today as part of Google Press Day. The service allows you to tap into Google's database of searches, to determine what's popular. For example, do a trends query on cars, and you can see the volume of queries over time, by city, regions, languages and so on.

Let's take a single search first and go through the motions. A query on ipod gives a chart going back through January 2004, which is as far back as Google Trends data goes. You can see spikes in searches, and these are often labeled with letters that lead to related news items. Google says it is using similar technology to do this as it does with company price charts in Google Finance.

Below the chart, you get some geographical and regional data. For example, you'll see most iPod searches are happening in New York, then in Irvine, then San Francisco, London and so on. That's the city data. Next is a Regional option, which gives you a breakdown by country (iPod searches are big in the UK then the US and Australia). Finally, you can narrow by language (Most searches for iPod are done in English, then Japanese).

Want to narrow in? You can do a variety of things. Using the drop down boxes, you can pick a particular month, such as last month. You can also pick a particular region, like last month just in the United States.

You aren't limited to single words. Enter multiple words by commas to do comparisons, such as google,yahoo,microsoft. That query shows you each term in a different color, and you can then see all the breakdowns for each word, as well. You can do up to five words in total. Want to do multiword queries? There's ways to do that -- check out the help page for more.

Sometimes when you do a search, you'll get something like this message:

Your terms - larry page - do not have enough search volume to show graphs.

What's happening here is that Google's working to help protect search privacy. There's a slight chance someone might enter something like their own name along with something embarrassing or private. Potentially, Google Trends could reveal this information.

My Private Searches Versus Personally Identifiable Searches article explains this issue more, and it's something Google used successfully to argue against handing over query data to the US Department Of Justice. Given this, it needed to put some protections into place. That mechanism is to only show data about queries that happen often.

"Something has to be in the hundreds of times per week for you to see trends," said Marissa Mayer, Google's vice president of search products & user experience, about the service. This is also touched on in the help page on the Google Trends site.

Some things to keep in mind. For example, Mayer cited to me a yankees,red socks comparison. Searches for Yankees are well above the Red Socks, so they must be more popular! Well, it's also a case that there are more people in New York than Boston, so there are more people potentially searching for the Yankees.

(Postscript: So I'm an idiot -- it's Red Sox, of course. And yankees,red sox for 2006 shows Red Sox actually much closer to Yankees. So cop-out time, the point in general remains valid. There are things that can skew the stats in ways you might not expect. For example, if you search for a particular company and you see growth in their name, are they more popular? In 2005, you might think so for Kryptonite. But go broader, you'll see a spike in 2004 associated with the Kryptonite locks-can-be-picked-by-ballpoint-pin-fiasco. That incident might have helped fuel some of the rise in following year -- searches that aren't necessarily reflecting a popular view of the company).

Another caveat. The geographic data is based on IP targeting, which isn't perfect. In particular, people who use AOL are often seen as if they are in Virginia, regardless of their true location.

How about query spam? Google's got a system designed to help filter for this, either if intentionally done or accidentally. For example, if it sees many queries all coming from the same IP address, that might be caught. Similarly, if it sees many queries coming from different cookies, it could be caused by the same person who rejects standing cookies. Each search would generate a new cookie, so potentially the same single person might be seen as different individuals.

"We are savvy to that case and make sure we saw queries from 100 different unique cookies that aren't fresh," she said.

Also, the data isn't filtered or consolidated in the way things happen in Google Zeitgeist or other search data mining tools. In other words, car brings back different results than cars. And if you want to see the dark underbelly of search, you can see in sex,ipod that if Apple sold a sexPod, it would leave iPod in the dust. You can also search for explicit adult terms, should you have the hankering.

Finally, Google rightly warns that this is more a play thing that something you can use for definitive predictions of popularity.

For a different spin on Google Trends, check out Barry's post, Fun With Google Trends. Now that we've warned you not to take the data too seriously, time for some comparisons anyway :)