RankBrain Google : The Definitive Guide
Google made a big announcement at the end of 2015. The new development of the company, RankBrain, began to take part in the processing of most searches. The sensational statement was worthy of a press conference at Bloomberg as RankBrain is an artificial intelligence that supports machine learning technology. Thus, the main difference between Google RankBrain and its predecessors is that the algorithm does not need to rely on humans to develop and improve its results.
According to official information, the main task of RankBrain is to determine search intentions for unusually formulated searches This category includes the search for expressions containing stop words, negatives, errors, and new formulations. These have all previously affected the quality of SERP significantly (read the related article – What is the SERP in 2020?). It is interesting to note that the share of completely new searches and formulations is about 15% of the total volume of searches.
Artificial Intelligence and Machine Learning
Artificial intelligence is the ability of an algorithm to imitate the cognitive processes of human thinking. Simply put, the ability to make decisions based on data. However, the database is incredibly complex in this case. Relationships and hierarchies are built on hundreds or even thousands of non-obvious factors of various importance and significance. The main difficulty of AI is the creative component of human nature. The emotional, aesthetic, and cultural value is difficult to transfer to the language of the algorithm.
To understand why artificial intelligence can be difficult to cope with the searches of millions of people, open your search history for last month. Imagine that you need to explain the connection between any two searches that you entered to a stranger without any context or knowledge about you. Now imagine that based on your explanation, this stranger must make an important decision. Therefore, information needs to be conveyed not just correctly but taking into account what decision a person should make based on it.
When we talk about machine learning, we mean the ability of an algorithm to update itself without human intervention. To better understand this concept, try to remember the last time you updated the operating system on the computer of one of your older relatives. The first step is to close all windows and programs. The second step is to disable everything. Only after that can you start the update. It is the same with algorithms. In order to update, you must first go offline. After that, changes are made to the algorithm, and a data packet is sent to it for processing. The processing results are evaluated by people and if after updating the indicators improve, the update is accepted. Most often, testing takes place in several stages. Only after evaluating the results, can a person return the already trained algorithm to the working order.
With machine learning, people do not interfere in the process, and the algorithm assesses the results before and after some changes, evaluates their usefulness, and decides whether to update or stay on the current version. The problem is that if the algorithm is entirely autonomous, at a certain stage its code will change so that no one can figure out how it actually works. This is a considerable risk in the case of search services since quite often it is necessary to repair and maintain something. In this case, the engineers would have to repair an unknown object in a dark room without instructions and tools. Therefore, a common practice is to use machine learning in isolation from the basic working algorithms that it is designed to improve.
When and Why Did They Create RankBrain
In 2011, a machine learning research project with the noteworthy name Google Brain was founded. The first development was the DistBelief system, aimed at improving the process of voice and text search. Interestingly, a year later, Google announced the launch of the Knowledge Graph project. The second version of DistBelief was Tensor Flow, where it had its own library added for learning the algorithm. In 2015, the Tensor Flow code became publicly available.
Together with these events, there was a struggle with search spam and low-quality content. Its main stages were:
- Panda update in 2010, aimed at improving search results by displaying the prioritization of sites with better and more useful content
- Penguin update in 2012, aimed at identifying unnatural link profiles and filtering such sites when ranking search results
- Hummingbird update in 2013, which affected 90% of the processed search queries, aimed at improving search results through semantic analysis, as well as analysis of synonyms and concepts
Learn more about algorithms in the Google Algorithms That Affect SEO article.
Thus, the development of machine learning systems and the fight against spam merged into a single process of improving search results. This necessitated the Google RankBrain update. Related article – What is SPAM and How To Protect Yourself.
Another major development was Google’s takeover of DeepMind in 2014. This is remarkable because DeepMind has developed the most advanced algorithm for the game of go, AlphaGo. This is a strategy game based on moving checkers in such a way in order to capture as much territory as possible. Its main difference from chess lies in the enormous variability of each game, which was the main obstacle in the creation of an AI capable of successfully playing Go. However, AlphaGo was more than just successful. The algorithm beat Grandmaster Lee Sedol 4 to 1, which was previously considered almost impossible.
Finally, the release of RankBrain itself was announced at a press conference for Bloomberg in late 2015, although the algorithm was launched somewhat earlier. When answering the question of what is Rankbrain, it is better to start from functions than from the definition itself. Its main goal is to improve the quality of search results by identifying similar queries, concepts, and analyzing user behavior. Through mathematical modeling, the Rankbrain algorithm translates natural language into vectors and graphs, representing words and searches as dots on a graph. The position and distance between the points indicate the degree of connection and its nature. But apart from that, according to Gary Elyes, the algorithm overcomes the stop word problem.
Thus, negatives, unfamiliar words, and simply strangely worded queries could no longer prevent the user from obtaining the desired answer.
Ranking Factors and Signals
At the moment, there are over 200 signals applied in Google ranking. The main three are links, content, and RankBrain. Each element is responsible for its own task and provides a foundation for the rest of the work. For example, a link profile indicates the nature of a site, its authority, major industries, and the areas of interest of its users. Content analysis allows you to determine the overall benefit of the material published by the site. Based on how often users read, share, and discuss what is published, it can be defined how valuable and relevant the content is to the audience. Then the question arises, what is Rank brain in SEO? AI solves the problem of matching by helping to connect pages and websites to searches as good as possible.
Since ranking results directly affect website performance, this subject has always had the most myths and misconceptions around. To check the presence of one or more determining factors in the formation of the SERP page, let’s conduct a small experiment. For example, let’s try to find vegetables that can be grown at home.
Note that the article with the least amount of text is ranked first. In addition, it offers the largest amount of unique multimedia content in the form of useful infographics and pictures. Chances are they are great at driving organic traffic from Pinterest. According to SimilarWeb, countryliving.com gets over 80% of its traffic from search, and 99% of that traffic is organic.
As we assumed, Pinterest is the main source of referrals from social networks.
Let’s see how often each of these resources posts publications in the corresponding section about home gardening.
Thus, the position of articles in search results is clearly not determined by the frequency of publication but by their quality, value for the audience, and the thematic focus of the site. It is important to take this into account when working on optimizing the structure of internal links. We should try to clearly indicate sections and topics, avoiding automatic and non-informative links without a subcategory
The last things we would like to draw to your attention are backlinks. Let’s check how many backlinks each article has and compare this with the organic ranking results. We need to open the desired section of SEMRush and enter the address of the article of interest in the search field.
The results may be somewhat surprising since the first positions are occupied by pages with a completely different number of backlinks.
However, it is clear that while backlinks to the content do not play a determining role in ranking, the authority of the domain and the sites linking to it can have a significant impact. A useful article on the topic is Build High-Quality Backlinks.
What to Pay Attention to and How to Optimize?
According to Google’s recommendation, content should first be optimized. The way you write can largely determine the ranking performance of your pages. Natural writing, the use of headings, meta descriptions, and structured data are just a few of the main aspects that can prove helpful.
Read more in the articles:
User analysis – total vs. returnees. How to calculate and what it affects (CTR)
Instead of telling you how to write, we’ll help you better identify opportunities to improve your content. You can start by looking at how often your users return to your site. This is a good indicator that can be used not only to assess the quality of work with the target audience but also to find directions for the growth and development of the project.
To do this, you will need to use the Google analytics service. Go to the behavior analysis section and select the “returning users” tab. To correctly calculate the number of returning users, set the time period of interest and base on the total number of site visitors during this period. Now, subtract the specified number of new visitors from the total number of users and you will get the most accurate picture of monthly website traffic. Also, these data can be prepared in different demographic sections, depending on gender, age, and other parameters. Each user group will show different dynamics, which will allow you to determine how effectively you are working with the target audience as accurately as possible. Important article on the topic – Guide to Google Analytics.
In addition, we recommend a cross-analysis of traffic data with the interests of users. As you know, three levels of interest are provided: general search queries, transactional intent, and other categories. With ready-made results on the demographics of new and returning users, you can get additional information about the interests of each segment, both broadly – which will allow you to create more relevant content, and in a narrow way – which will help you more accurately target PPC campaigns. More information in the article – Why Is Click-Through-Rate Important?
Content analysis – planning, properties, and distribution
As we said earlier, a lot of content isn’t always good. Likewise, quality content doesn’t always bring the highest conversions. Different people like different things and the more choices you provide, the more users will come back. Start with a list of competitors. How to identify your real competitors based on search traffic. We described this in detail in the SEO Competitive Analysis article.
Pay attention to a few key indicators:
- The presence of a blog, the average frequency of publications, their uniqueness, and length
- Social media profiles, followers number, engagement rate, an estimate of incoming social traffic, and other metrics
- Use of analytical tools, retargeting and mailing lists
- Availability of active advertising campaigns and main landing pages for them.
Thus you can clearly identify not only the main distribution channels of competitors’ content but also get a rough overview of their strategy. Based on traffic and user engagement data, you can filter out the top-performing options. Now all that remains is adapting them to your project.