Inside RankBrain: Changing SEO for the Automated Age

You probably heard the news about RankBrain – Bloomberg Business recently published an interview with Greg Corrado, a senior research scientist at Google, who described the technology as one of the engine’s newest and most sophisticated ranking signals.

Inside RankBrain

Considering it’s just one of hundreds of Google’s ranking signals, the details about RankBrain might not seem particularly important, but they are.

In fact, Corrado describes it as the third most important rank factor since its launch, and states that it’s already been affecting search results around the world for months.

RankBrain is an artificial intelligence (AI), developed by Google engineers as a rank factor that is now being used to process “a large fraction of search results every day.”

While little is known about RankBrain – except among Google’s scientists and engineers – some details have emerged that serve as subtle hints about how SEO is changing for the automated age.

The Latest Rank Factor

From what Corrado revealed, RankBrain should not be seen as a replacement for the Hummingbird algorithm, but as one of many factors that influences it.

Google is fairly secretive about its hundreds of rank factors, but it’s generally believed that they’re all somehow connected with Google’s perceived quality of the page.

Expanding on this concept, Search Engine Land has created a useful periodic table of the rank factors that they know something about:

search engine land periodic table

What RankBrain Does

Hummingbird still does its job of processing search’s and ranking pages based on the results that fit best.

RankBrain, on the other hand, helps out by offering various interpretations of what people search for, which ultimately influence the pages Hummingbird selects.

RankBrain does this by embedding the search terms into mathematical entities (what Google calls vectors).

Most search queries are pretty transparent, but when the AI stumbles across a search word or phrase that it isn’t familiar with, RankBrain makes a guess about the association those words might have with others.

These never-before-seen queries might include ambiguous searches or those using colloquial terms, and Google has long been working on ways to handle them.

In 2007, it reported that 20-25 percent of search queries were never seen before. Even now, it’s only reduced that number to 15%, which RankBrain is designed to help with.

Fifteen percent may seem like a lot, but it’s actually about 450 million of the 3 billion search queries Google receives every day.

According to Corrado, Google ran its own tests of RankBrain’s effectiveness, by recruiting their own search engineers – whose primary job is creating algorithms – to look at pages and guess what their related search terms might be.

The engineers were right an impressive 70% of the time, but RankBrain took the same test, and scored 80%.

Since its introduction as a ranking factor, RankBrain has become the third most important signal, but Google isn’t going into much more detail about it than that.

In fact, Google toyed with turning off RankBrain in a few experiments, and found that the results “would be as damaging to users as forgetting to serve half the pages on Wikipedia,” according to Corrado.

Learning the Meaning Behind Words

What sets RankBrain apart from all other rank factors deployed to date is its ability to learn as it goes along, whereas all other factors so far have been based on information and insights that search engineers have retrieved.

As it turns out, Google has been investing in AI – specifically learning systems – for a long time.

Although not named RankBrain at the time, Google blogged about the technology back in 2013, describing how deep learning technology has helped them with image classification and speech identification.

They’ve since applied the same technology to natural language, and began promoting their research with word2vec, a tool that calculates the vector representations of words that can be used for further research or with language processing applications.

In the blog post linked to above, Google offered an example of the technology’s ability to learn the relationships between words.

If the AI already knows that the relationship between Paris and France is the same as the relationship between Berlin and Germany, but not the same as the relationship between Madrid and Italy, it can identify more capital and country connections by simply reading news articles:

1
In repeated tests, the AI managed to put similar countries next to each other, and also made a parallel arrangement of the capital cities, with no specific instructions or information beforehand.

How It Works for Search

It’s too early to dream up my own examples of the kind of searches that might show up on RankBrain’s radar, but Google has offered Bloomberg Business and Search Engine Land a few of their own.

Search Term: Barack

According to Corrado’s interview, searching Google for “Barack” may have only returned results for pages that contain this exact word in past Google algorithms.

But with RankBrain’s effort to identify the lexical, physical and social relationships between words, a search for “Barack” will now return pages that match search terms closely related to it, such as “US President,” “Barack Obama,” or “Michelle Obama’s husband.”

barack obama

Search Term: What’s the title of the consumer at the highest level of a food chain?

This is an example of what Google considers to be an ambiguous search query. I agree – I initially didn’t understand what was being asked.

Most people think of a “consumer” as someone who purchases things, but it’s also a scientific term used to describe individual players in nature, a synonym of “predator.”

My search for “What’s the title of the consumer at the highest level of a food chain” shows RankBrain’s work in action:

RankBrain

The search returns relevant results that include the words “predator” and “consumer,” demonstrating RankBrain’s ability to identify the relationship between these words and apply it to returning the right kind of results.

Search Term: How many tablespoons in a cup?

Search Engine Land pressed Google for another example of an ambiguous search that RankBrain could help with. They responded with, “How many tablespoons in a cup?”

This search would be considered ambiguous because, while Australia and the United States both use tablespoons and cups as forms of measurement, their definition of each is different.

To clarify things, Google said that RankBrain can now identify which results are relevant for people searching in Australia versus the United States.

It’s an interesting example of one of the many situations where keyword use is downplayed. Even if you have a page highly optimized for these keywords, if you’re referring to US measurements, Australian searchers likely won’t even see the page.

What to Expect

Now that Google has revealed a bit about RankBrain and how important it is for search, many marketers are already worrying about how this new ranking factor might affect their web traffic (as they have with every other algorithm update).

Of course, it’s important to pay attention to a change as big as this one, and I’ll definitely be learning all I can about it for my own work at Louder as more details emerge.

Still, there’s no need to hold your breath in fear of imminent rank penalties to your website. Google snuck this one up on everyone – RankBrain has already been affecting search since sometime early this year. If your traffic was going to tank, it would have already.

Beyond penalties, though, many want to know what they can do to leverage RankBrain and keep Google happy with their website.

At this point, this AI technology is so new that it’s hard to say exactly what strategy you should take.

And since RankBrain has the unique ability to learn and improve over time – unlike any other factor – what it considers when ranking your web pages will continue to grow, change, and become more sophisticated.

Since RankBrain’s main job is to better interpret queries, pages, and user intent, I think it’s safe to say that this ranking factor will mostly serve to help online marketers.

You don’t want someone clicking on your website from a Google search, only to find that it’s not what they’re looking for and bouncing.

Your goal as a marketer is to draw in the correct audience, and RankBrain is trying to connect users with the pages they’re actually looking for.

It may seem simplistic, but focusing on quality content that delivers something valuable to your target audience is the main thing you need in order to work well with RankBrain now and in the future – advice Google has been giving all along.

Do you agree or disagree with my interpretation of RankBrain? I’d love to hear your thoughts, so share them by leaving a comment below:

Note: The opinions expressed in this article are the views of the author, and not necessarily the views of Caphyon, its staff, or its partners.

Author: Aaron Agius

Aaron Agius is an experienced search, content and social marketer. He has worked with some of the world’s largest and most recognized brands to build their online presence. See more from Aaron at Louder Online, their Blog, Facebook, Twitter, Google+ and LinkedIn.

CTA image

Free Google Algorithm Changes Tool

Find out in real time how Google’s algo updates affect rankings and visibility. No login required!

Show me this tool

2 thoughts on “Inside RankBrain: Changing SEO for the Automated Age”

  1. Great post! Very informative. I wonder if there is a way to reverse engineer the word2vec program in order to project success rates and related content matrices for an article or blog post.

  2. Great article Aaron. I’ll be interested to see if RankBrain is going to take into account and track personal searches conducted when the user is logged into their gmail account to create even more tailored results.

Leave a Reply

Your email address will not be published. Required fields are marked *