RankBrain: An investigation of a Search AI

Posted by on Apr 10, 2016 in Algo Changes, Google Updates


It sounds like some snake oil concoction an SEO thought up to help sell the local mom and pop shops on SEO. But, it’s not. It is actually one of the top ranking factors for Google. It’s existence was announced in October of last year, although the patent was submitted in 2012.

So, what exactly is “RankBrain” and what does it do?

RankBrain is officially considered an artificial intelligence system. Reason being is that it “naturally” evolves; growing in “intelligence” as time passes and it accepts search data. What search data you say? Your search data. Online search is considered natural and organic when users perform search actions without other input and as a way to find answers to their questions (queries). RankBrain’s intelligence is focused on evaluating the way organic search is being used, semantic web activity, events, and much more. So there are no worries about this AI developing sentience and enslaving us all. RankBrain is an artificial intelligence program, not a ‘true’ AI in the traditional Terminator or I, Robot sense.

Logic, its so difficult.

According to the actual patent, RankBrain focuses on contextual matching and substitution data to help improve the search results. What this means to the layman is that Google search results will “seem” more intelligent as it produces results that may include organic (and some paid results) that may not align directly to the query. Now, that doesn’t mean RankBrain replaces search results; it is one of the more than 200 search ranking signals used by the search engine. Even more so, it is one of the top three. The other two most important Google ranking factors? Content (duh) and links (double duh). Search Engine Watch pointed out that RankBrain fills in a need that exists around live updates.

Content is and has always been and will likely going to always be, king. Links provide signals of importance, authority, and credibility of said content. But, what about live actions and search activity in connection to the content and links? Hello RankBrain! 15% of all queries completed each day are new. Furthermore, Search Engine Land outlined that Google processes 3 billion searches each day. With that and the 15% means 450 million entirely unique searches are created daily!

Forbes‘ contributor and SEO Jayson DeMers offered a great outline of what RankBrain is NOT:

  1. Not the ‘new’ Knowledge Graph. (KG is a AIP too, but focused on offering definitions and details on the query)
  2. This isn’t an algorithm change or update.
  3. RankBrain isn’t a new type of search. It impacts what shows in the results.
  4. RankBrain isn’t a robot. It is really just mathematical equations working together. I think of it like compound interest, as time goes on and the calculations are completed, the completed calculations impact all future computations.

Will there be more announced algo changes?

In the past, Google would have to internally test updates, develop the change, then push the change live. This is where we got some of our fun names like Penguin, Panda, and Hummingbird to name the latest. Those days are over (or deprecated for the techies in the crowd). Now, if there are any changes to the algorithm, they will likely come in the form of live updates or “bolt on’s” to the current algorithm. That isn’t to say Google won’t update the core like when they switched to “Caffeine” in 2010 and “Hummingbird” in 2014.

The Wrapper

You’ve probably done a search or two since RankBrain was released on Google. Any changes to the SERP aren’t being completed by some robot locked in a closet. Design isn’t RankBrain’s forte, search is. Verve Developments offered some great examples of what RankBrain “physically” does to search. Part of the program focuses on vectors which help align contextual targets:

vector(‘Paris’) – vector(‘France’) + vector(‘Italy’) results in a vector that is very close to vector(‘Rome’ )

vector(‘king’) – vector(‘man’) + vector(‘woman’) is close to vector(‘queen’)

And the vector (san francisco) would have close vector relations with (los_angeles) (golden_gate) (california) (oakland) (san_diego)

In another example, vec(“Russia”) + vec(“river”) is close to vec(“Volga River”), and vec(“Germany”) + vec(“capital”) is close to vec(“Berlin”).

Make sure to check them out and David Harry’s post to dive deeper into the Google patent.

As things with the search giant get better, that can only mean that things for all of us will get better, right? We can only hope that Google makes sure to include kill switches into any of these artificial intelligent programs or at least some secret backdoors; they can check with Apple on that option.

A. Chris TurnerAbout the Author: Chris Turner is also known as ChocolateSEO. CSEO is Chris' Nashville search marketing and consulting service offering a variety of services to help you, your company and any website maximize web-based marketing opportunities. He is the father of three girls, one boy (finally) and husband to the wonderful Savannah. Join the author's circle: Chris Turner on Google+.