Back in October 2015, the biggest search engine announced their latest algorithms.
It was named Google RankBrain and became an integral part of the biggest search engine.
The story was first presented in a Bloomberg article.
RankBrain is one of the newest additions to Google’s algorithm (but not an algorithmic update).
It represents a machine learning system that should help Google understand search queries better.
So far, Google has provided us with very limited information regarding RankBrain.
Due to this fact, we have been left with a lot of room for speculation.
Here are some facts and presumptions when it comes to this revolutionary system.
What is Google RankBrain?
This is a machine learning system that should provide users more relevant results to their queries.
Google RankBrain is an automated system. It is able to learn by itself without any human interference.
Furthermore, a great thing about it is that it can learn from its mistakes and continuously improve its results.
Have in mind that this wasn’t Google’s first project involving machine learning. Google news is based on the same concept.
By noticing patterns and relationships, it is able to conclude that different stories are on the same topic.
RankBrain can help us with numerous things such as ambiguous queries, phrases with numerous meanings and slang.
In short, with everything that Google previously had issues with.
In that regard, it was a much needed improvement of the search engine.
To top it all off, RankBrain can follow our train of thoughts and learn from our previous queries.
How does Google RankBrain learn?
Each day, about 15% of all Google queries have never been seen before.
This amounts to whopping 500 million queries.
In these situations, RankBrain is able to make an educated guess based on understanding of the language.
However, it doesn’t stop there.
If Google provides users with bad results, it can easily be determine this. How?
If these stats are low, RankBrain will give another set of result the next time such a query is typed in.
Before the introduction of this algorithm, Google had to create synonym lists and establish connections between entities. As you can presume, this took a lot of time.
Again, there are 500 million new searches each and every day.
Have in mind that a vast amount of these queries are long-tail keywords (not medium tails or head terms).
They are sometimes ambiguous in nature. If we take this into account, the problem becomes even bigger.
This machine learning system is able to establish patterns and relationships between different entities.
However, this is not all.
After a successful query, it memorizes the results and store this information for further use.
As a result, it is constantly and exponentially able to improve itself.
How did it all start?
Although general public heard about Google RankBrain in October 2015, it was first introduced at the beginning of the same year.
The entire process was performed in several stages. This allowed Google team to monitor progress as they went.
Most people would presume that RankBrain is an online system.
This is completely wrong.
All of the learning processes are performed offline.
The procedure is performed during two major stages:
- RankBrain takes historical data and base its initial prediction on it
- These predictions are additionally tested before they can be presented to the final user
These two stages are redone each time a person makes a unique query.
Machine Learning vs Artificial Intelligence
There are many misconceptions regarding machine learning and artificial intelligence.
Most people think that they are synonyms but this cannot be farther from the truth.
Machine learning is a process during which computer system is able to improve over the course of time.
Unlike other systems, it is completely autonomous. It doesn’t require updates or human interference.
Artificial intelligence is the final step of the process.
When machine learning becomes advanced enough, it can be regarded as AI.
In other words, not requiring any additional improvement. At this point, artificial intelligence is still a hypothetical term.
Google RankBrain is not the only system that uses machine learning. Other examples include email filtering, photo recognition, etc.
At this point in time, artificial intelligence is still far away. Even machine learning has a lot of issues as it is.
This is still a new area for all of us and we have just starting to explore the possibilities.
Connection to Hummingbird
There are several speculations when it comes to Hummingbird and RankBrain.
According to some experts, Google RankBrain is a part of Hummingbird algorithm (SEMRush).
Others believe that this algorithm is in fact an autonomous modification that should help out Hummingbird (Moz).
For example, when user enters a query, it will be first handled by RankBrain and then processed by Hummingbird.
In that regard, both of them have their task and in conjunction. They must provide optimal results.
Machine learning is nothing new to Google. They already had some similar projects in the past.
In 2013 Google introduced deep learning technology for speech identification and image classification.
With this in mind, it shouldn’t surprise us if RankBrain was actually based on Hummingbird technology (or at least some part of it).
Both of these projects may suggest that Google is slowly going towards voice search for mobile devices.
So far, RankBrain has shown promising results. I’ll talk more about them later, so keep reading.
This is to be expected given how many people access internet through their mobile devices.
Sooner or later, this issue will have to be addressed properly.
Anyway you put it, one thing is for certain. They are both important for understanding meaning behind words and reading user intent.
Connection to Google AdWords
Apparently, relevancy is the main factor that Google RankBrain takes into consideration.
What happens when we enter a search query in Google?
RankBrain will analyze numerous pages and give them score based on their relevancy.
This score will range from low score for irrelevant to high score for very relevant content.
Google AdWords uses a similar concept. It has Google AdWords’ Quality Score which determines relevancy of a page.
This technology doesn’t require external signals such as links to determine how relevant the content truly is.
This poses certain questions such as, whether links will be an SEO factor in future.
If this happens to be the case, Google will eventually create a system that will completely rely on visitors and their feedback regarding quality of a page.
Like most things, this can lead to negative and positive issues.
In case that RankBrain technology takes over all rankings, it means that bloggers will have no impact on SERP thus eliminating any negative methods and reliance on links.
But, at the same time, this means that laymen will have a bigger say that experts themselves possibly leading to degradation of content quality in first positions of SERP.
Only Google knows what will happen in the future.
Nevertheless, at this point in time, it seems that both systems will have certain cracks that can be exploited by SEO professionals.
How does RankBrain work in practice?
Google RankBrain works by embedding words into vectors.
Naturally, this transformation is necessary given that the system cannot understand words but it can understand vectors.
But, in order for the entire system to work, this is not enough.
RankBrain also has to recognize differences and similarities between different vectors. This way, it can connect the dots and understand user intent much better.
Here is an example from bruceclay.com:
Search query used was “What’s the weather going to be like in California?”
As you can see, “California” has the dominant position.
All the other related words are placed within the same space. Distance represents relevancy.
Those that are closer to each other will be more relevant and vice versa. Articles will gain relevancy based on a number of highly related keywords.
This technology is probably not new. In fact, RankBrain have big similarities with Word2Vec.
Keep in mind that same experts worked on both of these projects, we can easily conclude that RankBrain is at least modification of the previous system.
However, it is hard to tell how similar they actually are.
Word2Vec and RankBrain
Word2Vec takes a bag of words and turns them into vectors. This is the basic ingredient of semantic system.
It was built upon two different models: skip-gram and continuous bag of words.
While continuous bag of words is used to predict current word based on the words that are surrounding it, skip-gram is doing the opposite, predicting neighboring words based on the main word.
Basically, Google RankBrain is learning through process of trials and errors.
Based on user’s clicks and their satisfaction with provided results, certain pages will improve (or degrade) in rankings.
System is able to constantly provide individual with more and more relevant results to their queries.
In that regard, consumer is put in the center of the process.
At this point, it is only used for some queries (those that Google cannot provide definitive results for).
However, this doesn’t mean that RankBrain won’t take over the other results as the time goes by.
What’s the technology behind RankBrain?
Nevertheless, even with all our knowledge and educated guesses, we are still not quite sure what is the precise technology behind it.
Experts speculate that there is a ranking mechanism embedded within the system allowing it to rank different pages afterwards presenting them to user.
Most likely, RankBrain has a lot of similarities with Google’s previous algorithms.
Naturally, the guys from Google took it a step further.
So far, RankBrain is the best system when it comes to understanding content and dissecting a page.
Some may speculate that RankBrain is mix of ranking signals and a search processing tool. At this point, it is hard to say.
Whatever it may be, RankBrain is the most advanced piece of Google’s arsenal and as such, it will probably remain in place for some time.
Also, in future, it may be used as the starting point for creation of new technologies.
The third most important ranking factor
When ranking websites and content, Google relies on more than 200 different search signals.
According to Google, RankBrain is already the third most important ranking signal. However, there is still debate as to how this impact is perceived.
The difference between RankBrain and all other ranking factors is its ability to learn on the move.
Other factors depended on second party and its input, in other words, all the information procured by search engineers.
Most experts believe that RankBrain isn’t similar to other traditional ranking signals.
Furthermore, we presume that it in not exactly a direct ranking signal.
Instead, it is something that affects queries and through them, Google search as a whole.
Nevertheless, it shouldn’t make an enormous impact on total search results.
Why was RankBrain invented?
Initially, it was created as a way to deal with brand new searches. But, as it turns out, its biggest potential is to deal with ambiguous, never seen before or strange queries.
In that regard, if you already have a good position for a particular keyword, RankBrain most likely will not have any impact on the search.
In fact, this machine learning system is best when dealing with queries that lack data.
If the keyword is competitive, with enough monthly searches, Google will most likely have adequate suggestions by now.
This being said, RankBrain can only assist by sending user to most relevant results in case that search is not clearly defined.
But, this is where the problem lies.
How can we adapt to RankBrain?
There is no point in adapting to Google RankBrain.
These troublesome queries usually do not have enough volume to justify optimization.
Having this in mind, you should continue optimizing your site for your basic, simpler queries.
For example, if you type “What is the best animal?” Google can send you to pages about dogs (man’s best friend), lions (king of the jungle), monkeys (smartest animals) etc.
After series of trials and errors, Google will conclude that the proper answer is dogs.
So, if you have a nice, well-optimized page on this topic, Google will most likely reroute user to your page.
Having this in mind, you won’t have to optimize around keyword “What is the best animal?”
The future of SEO and optimizing for RankBrain
As previously mentioned, normal queries probably won’t be affected by the RankBrain.
Of course, that doesn’t mean that search won’t change in future. Who knows; external factors such as link may become completely useless in future (even though this is really far-fetched at this point).
If this happens, Google search may completely rely on RankBrain’s judgment.
But, let’s not concern ourselves with things that may happen.
If you wish to get the most benefit from this update, you will have to focus on on-page elements.
In order to harness all these ambiguous queries and rank well for them, you will have to look relevant to Google.
Now, as previously mentioned, there is no point in optimizing around strange keywords or slang.
However, you can still try and rank for shorter and more direct keywords.
According to our presumptions, relevancy of the page will be determined by click through rate, bounce rate and time spent on a web page.
So, if Google makes an assumption and send user to your page, make sure that it is well optimized and that it can attract his attention.
First, you will have to have a nice title and a well-written META description.
Remember, if your query is completely irrelevant to users, they won’t click no matter what.
However, if it peaks their attention and is related to their query, they might get intrigued.
By providing a nice META description and title, you can subconsciously persuade visitors to give advantage to your copy.
This will ensure that your copy get good CTR.
But that is not all!
Is your content REALLY good?
After that, you will have to make sure that your content is good enough.
Even if you draw a visitor to your page, it doesn’t necessarily mean that they will like the provided information.
If you manage to cover everything and provide fresh data, visitors will most likely appreciate it and in return, they will spend some time reading the content.
This will send positive signals to Google. In terms, this may reduce bounce rate at the same time.
Have in mind that these are only speculations.
Based on things that we know about RankBrain, this is the most likely turn of events.
Nevertheless, the quality of your content as well as good on-page SEO are already crucial parts of search engine optimization.
In that regard, your strategy shouldn’t change too much in comparison to what you are currently doing.
However you put it, users are the ones that benefit the most from this update.
Overall quality of the search has definitely improved which, in the end, serves to improve user experience.
How to write content for RankBrain
Even though I am sure that you get the picture when it comes to on-page optimization, here are some additional tips that can give you a better insight.
Keywords are not enough
This is not a new concept. For some time now, Google is focusing on semantic search and overall meaning behind the text. With Google RankBrain, this will become even more emphasized.
Having this in mind, optimization will no longer revolve around putting the exact words and phrases in text. Instead, they will have to provide an answer to a question.
Therefore, SEO experts and webmasters will have to focus on their audience, products and all the things that may be the true intent of user’s search query.
Conversational tone of the content
Previously, users had to modify their queries while looking for keywords that will trigger the proper response providing them with useful information.
With Google RankBrain, this is no longer the case.
Instead of adapting to the search engine, search engine will have to adapt to users. This means that queries will become more natural and direct.
In that regard, Google will become more of a medium between two entities.
User that asks a normal, common question, formulated for another human being and content that talks directly to user and provides step by step explanation of the process.
Google always emphasized that providing high quality content is the main focus of their search engine. Information procured from authoritative websites and extracted from different studies, is crucial.
With this in mind, RankBrain will most likely give advantage to content that quotes important resources, that is backed by scientific research and has all the relevant data on the topic.
After reading your article, visitor should fully understand the topic, the same way experts in the field see it.
Importance of Co-occurrence
Al Gomez puts co-occurrence high on the list of priorities.
What does this mean?
Whenever you write an article, certain terms or related group of terms should have high frequency.
If there are multiple synonyms or related keywords that are more likely to appear in such a text, Google will give content higher priority.
Let us use this example:
This is an article written by QuickSprout.
It is the first result for “boosting sales using SEO” query. As you can see, there are a lot of different related keywords that are making article that much more relevant to Google.
Usually, if you wish to have high co-occurrence, you will have to have a longer piece.
This way, you can fit all the required words and phrases without looking unnatural. Long content is king!
How did RankBrain impact Google search results?
The big question is: Has RankBrain actually improved Google results?
How did it handle all these queries which previously caused so much trouble?
In a small study performed by Google, company used their own engineers to assess pages of different websites.
Their task was to find all the related search terms.
Have in mind that these are experienced professionals who create algorithms for the company.
While engineers had 70% efficiency, RankBrain had 80%.
Google went one step further. On several occasions, they even turned off RankBrain during their experiments.
According to them, if they were to turn off the system, it would be as damaging to users as forgetting to serve half the pages on Wikipedia.
Eric Enge’s case study
Some time ago, Eric Enge and team from Stone Temple Consulting have performed a study on Google’s rich answers that included a database with 1.4 million results.
They took a snapshot of these results.
Recently, after Google introduced RankBrain, they took yet another snapshot.
They wanted to conclude whether Google understood some queries better after implementation of RankBrain.
In the process they excluded all the results that weren’t put together properly as well as those results for which there is no proper answer.
Stone Temple Consulting made a really interesting discovery.
According to their study, results improved by 54.6 %. You will probably agree that this is a significant difference.
This is yet another interesting RankBrain before and after example:
As you can see, Baseline Set gave us some unrelated results which had parts of the keyword within them.
However, after RankBrain, Google is able to understand that term “weak” relates to security.
Thus, through understanding of user intent, it can provide user with much more relevant results.
Searchmetrics also performed a study of their own.
According to it, high relevance is crucial for optimization and it is aligned with user intent.
Page has to provide good user experience, to be properly structured and to provide other related information which may answer all additional questions user may have.
Main findings about Google RankBrain:
- Depending on keywords used, top ranking factors may be different
- No matter what, relevance is crucial for rankings
- Relevance score is more important than other ranking factors
In cases where Google already has a relevant list of results, RankBrain is not used.
When we talk about individual examples, RankBrain had an amazing impact on keywords such as “the”, “graphic”, “without”.
According to Eric Enge, these words were very problematic for Google prior to this machine system.
After RankBrain, search engine is able to process them properly.
But, this is only the tip of the iceberg. There were some other improvements besides mentioned keywords when it comes to understanding of natural language.
Have in mind that Google’s algorithms are constantly improving.
In that regard, it is hard to say if these changes are directly influenced by RankBrain or something else.
However, if we had to bet, there are high chances that Google RankBrain had direct impact on improvement of results.
Examples of RankBrain in action
Google RankBrain is excellent when it comes to understanding user intent.
Previously, SEO experts had to optimize around their keyword.
When a person enters a keywords into Google search bar, search engine will present them with pages that contain one of more words belonging to the mentioned keyword.
However, this doesn’t necessarily mean that they will get proper suggestions.
Let us start with the keyword “Barack”.
Previously, you had to enter exact match keyword to get information on Barack Obama.
After RankBrain, you can enter phrases such as “US President” or “Michelle Obama’s husband”.
By using stored data, system is able to determine topical and semantic relationships and return proper answer even if the question is not direct.
But wait, there is more…
Does RankBrain follow our train of thoughts?
RankBrain is great at following your train of thoughts.
For example, type in “Barack Obama”. After that, try asking Google “How old is his wife”?
Machine learning system will use your history and semantics to determine what is the real question.
After that, he will show you picture of Michelle Obama with her age included.
But, RankBrain is also great with voice queries, which is to be expected given that Google already experimented with deep learning technology and speech identification.
However, due to their nature, voice queries are more likely to yield bad results.
In this example, query “the name of the fat guy with the beard in the movie with the four guys in Vegas?” was used.
As you can presume, this would be a pretty hard task for Google back in the day.
Search engine would most likely cut the phrase in different words and return results that have some of these words present on the page.
After RankBrain, Google understand the query and is able to determine that user is searching for “The Hangover”.
Another example would be “What’s the title of the consumer at the highest level of a food chain”.
Obviously, this is a very ambiguous query.
Even if you are not a robot, it will take some time to understand the question.
In this example “consumer” is used to describe predators.
Back in the day, Google would most likely focus on “consumer” as a word, not being able to understand that we are looking for predators.
Thus, we would likely get results suggesting shopping malls or something related to food production.
Google was once again on the money.
Without any doubt, Google RankBrain shows us things to come.
Google has always tried to help final users, improving search result and proving them the most valuable information in the process.
In that regard, RankBrain is one of the important tools that will improve their experience.
Similarly, this looks like an important step ahead for machine learning.
Most people are looking forward to the idea of artificial intelligence.
Question remains whether we will reach this point due to many limitations and prerequisites that have to be accomplished.
Anyway, congratulations to Google for being able to incorporate different technologies and constantly improve systems.
Google RankBrain is certainly a much needed improvement.
The potential of this system looks limitless at this point. It is definitely a new and exciting chapter in Google’s history.
If this system becomes a dominant element of Google search, Google users will become those that are determining the rankings (again, it’s a bit far-fetched to me).
Nevertheless, this doesn’t mean that SEO will vanish as a profession. It will most likely have to put stronger emphasis on on-page elements.
How does RankBrain affect your SEO and content marketing efforts? Share it with me on Facebook, Twitter or some of the other social media channels. I’d love to know more 🙂