Over the last few weeks there has been a lot of fuss over Google’s new machine learning algorithm. The ‘Rankbrain” algorithm is Google’s attempt at artificial intelligence. Greg Corrado, one of Google’s senior research scientists, said that “a large fraction” of the millions of queries that come through Google every second have been interpreted by the RankBrain system. RankBrain itself is not actually new – it is, in fact, the third most important signal that contributes to the results of your queries. It works using a machine learning system which guesses what a user wants based on what they are typing. If you ever enter an ambiguous query into the search engine – something that has never been submitted before or that could have more than one meaning, then RankBrain will handle it. Believe it or not, brand-new queries account for 15 percent of all searches on a daily basis – which means that there are about 450 million queries per day that are unique in their nature. This is a surprising statistic when you consider that there are three billion searches per day. RankBrain is performing very well at the moment. When Google asked its own engineers to predict which pages would rank at the top of the search results for a given query, the engineers predicted the rankings correctly for 70 percent of the queries – while RankBrain predicted it correctly for 80 percent of the queries, even though it is not the only signal that matters.
Google is constantly revisiting the way that it ranks websites, and over time different factors are becoming increasingly important. What matters today could be entirely different to what matters tomorrow. Corrado himself admits that RankBrain has been a far more successful experiment than he first predicted, but it is just one part of the algorithm. Google Hummingbird saw the Google ranking algorithm massively rewritten, and there are quite literally hundreds of small tweaks and adjustments made to different parts of the search algorithm every year as anti-spam, content quality, mobile and local factors are considered. RankBrain is proof that machines can do something that looks a lot like learning – and it shows that they can interpret what a human is looking for quite well. They may not be able to ‘think’ in the traditional sense but you cannot fool the system with simple tricks anymore, because rankings are done by something far more sophisticated than the average software.