Google’s search ranking algorithm has always been very mysterious. People in the field of Search Engine Optimization, or SEO for short, have been trying to reverse engineer its inner workings for years, with only limited success. Most people do not know the first thing about Google’s search algorithm, and it turns out there is more to know than most would assume. However, recently Google has spoken a little more open about their program Rank Brain, which is a ranking algorithm powered by machine learning, artificial intelligence.
So what is Rank Brain? Rank Brain is, effectively, the artificial intelligence that decides in which order results should be shown when a Google search is performed. When someone makes a search query, Google’s search algorithm trawls through the data it has on millions of websites to decide which ones are most relevant. Then, Rank Brain’s job is to decide which of the results is most likely to be clicked on, and present that one first. It is powered by machine learning, which is a technology that has been exploding lately.
Machine learning is a way to teach computer programs to learn things themselves through experience, as humans do, rather than needing every parameter to be manually defined by a programmer. This allows a powerful enough machine learning program to surpass humans in fields like prediction of behavior, given enough data. Rank Brain has been learning for over a decade which links are clicked in response to which searches.
Each time a search is performed, it guesses which link will be clicked, and puts that one first. If that link is clicked, Rank Brain gets a reward function that tells it to keep on guessing in that way. If it is incorrect, Rank Brain takes into account what link was actually clicked, and guesses more accurately the next time.
What is Rank Brain’s purpose, you may ask? The idea behind a search engine is to get people to websites relevant to the keywords as quickly as possible, so why not just show the most relevant results based on keyword density and other objective measures? The idea behind a learning machine is that it can progress beyond hard data and numbers, which is something computers have never been able to accomplish before.
Computers that have undergone machine learning algorithms display something akin to human intuition. Rank Brain does not have a solid reason for suggesting a website first, it just remembers thousands of previous searches and uses that experience to generate the computer equivalent of a gut feeling on what you want to see.
The amazing thing is that Rank Brain works in a way that a conventional program never could. The amount of coding that would be required to take into account nearly two decades of data is simply not possible to be done by a human team. There is too much noise in that information to wrestle it into useful code for a computer program.
The genius of Rank Brain is that it writes itself as it goes, learning from its own mistakes and editing its own code to do better in the future. While machine learning algorithms are less rigid and predictable than traditional code, they are also much more powerful and flexible.
So what is Rank Brain’s place in Google’s overall search engine algorithm? The Google search algorithm, called Hummingbird, is a very large program made of individual pieces. Rank Brain is just one piece in the puzzle, the one that is responsible for ranking the search results. When a search query is made, Hummingbird is responsible for everything from parsing the query, to looking through data, to presenting relevant results, to filtering out spam from those results, to ranking the results in order of importance, and finally to presenting the results to the person who asked the query, all in under a second. So how many ranking signals are there?
Besides Rank Brain, Hummingbird is made up of a number of other parts with specific jobs to do. SEO experts will be familiar with Panda, Penguin, and Payday, all of which are designed to reduce spam from search results. Each of these programs are designed to independently review website quality to ensure that sites that are simply keyword spam do not make top spots. SEO is a delicate dance between being relevant enough to top the charts while not triggering these anti-spam filters, which is why proper keyword density is so important.
In addition to anti-spam routines, there are a few other components of Hummingbird. Pigeon is a subroutine that uses the user’s location based on mobile GPS or IP address to promote search results that are geographically relevant. This is why when a search is done for something like a chain store, Google is likely to present a map with the location of the nearest outlet of that chain.
Top Heavy is an algorithm designed to bring down the search ranking of sites with too many advertisements. Mobile Friendly checks the usability of sites on mobile devices, and raises the search relevance of those with the best usability. Pirate is designed to filter out results that are involved in illegal behavior related to copyright infringement.
So, Hummingbird is not so much an algorithm as it is a large pile of programs all running in tandem. When a search is performed on Google, all of these programs, and probably even more that are not known by the public, do their jobs to filter, refine, and rank search results to display only those that are most likely to be relevant to the consumer. Rank Brain is one part of this large engine.
So what is Rank Brain? Rank Brain is a learning AI that controls the order in which search results are shown on Google. It uses over a decade worth of search data to present what it believes to be the most likely pages to generate clicks first in the ranks, using its years of experience to generate the AI equivalent of a gut feeling.