On average, there are more than 3.5 billion Google searches every day, and a small percentage of these are requests that have never been made before. In an attempt to handle these obscure and hard-to-find searches, Google has developed an artificial intelligence system called RankBrain. According to Google, RankBrain tackles a “very large fraction” of all the total number of Google searches, which are about 15% of the millions of queries it receives every second.
Google searches are ranked based on ‘hundreds’ of signals including location, key words, the site’s ranking and more. Its current technology relies on discoveries and insights that people in information retrieval have had. RankBrain instead uses AI to turn words into so-called ‘vectors’ the computer can understand more easily. If it sees a word it doesn’t recognize, it takes an educated guess at what it could mean, based on other words and phrases that may have a similar meaning.
The system then filters the results and presents the most appropriate links to the person making the search. Every time it makes these guesses, it monitors how the person making the searches responds to the results and can adjust its filtering process accordingly.
Google says it’s only used RankBrain to handle this massive search load for the past few months, but Google says RankBrain is actually better at predicting top search results than their own search engineers.