I work on a team that develops autocomplete—the feature that provides predicted searches while you type. When you combine autocomplete with Google Instant, you can really accelerate your searching. Because it is so important to your search experience, we’ve been looking for ways to provide predictions for even more queries. Today we’re improving the predictive powers of autocomplete, helping you search for things even when no one else in the world has.

One of the main ways autocomplete works is by looking at the most popular searches on Google. For example, most people who type “w-e-a” are going to search for [weather], so Google can make that prediction. What’s tricky is that a huge percentage of the queries we get have almost never been typed before, so this makes it difficult to provide predictions based on popularity. For example, very few people have searched for [florida state senate building], so until today, even when you typed nearly the whole search query, you wouldn’t get a prediction.

Now what we’re doing is making predictions based on only part of your search—specifically, the last word or words. While few people have searched for [florida state senate building], many more have searched for [state senate building]. By looking at just the last part of what you’ve typed into the box, in this case “state senate bui,” we can generate a prediction for “building.” You’ll see a dropdown box below the end of your search with predictions for just that word.