SEMRush has made an attempt to understand how RankBrain works and if there is anything that can be done to optimize and prepare your site for meeting it. Read more here!
RankBrain is a system developed by Google that uses machine learning and artificial intelligence to improve search results and interpret new search requests, that is, terms that have not been searched before. Some experts consider RankBrain to be part of the Hummingbird search algorithm.
Hardly anyone in the digital industry hasn’t heard of RankBrain. It is no coincidence that this system, in addition to having an immense impact on search results, causes many questions and discussions.
We made an attempt to understand how RankBrain works and if there is anything that can be done to optimize and prepare the site for meeting it. Let’s start at the beginning.
What is RankBrain?
Some experts consider RankBrain to be part of the Hummingbird search algorithm, which was launched to help Google better understand the meaning of search requests expressed by exact keywords.
The RankBrain was first mentioned on October 26, 2015, by Greg Corrado, senior research analyst Google. This is how Greg explained the operating principle of the algorithm:
“If you’re searching for an ambiguous phrase, using colloquial terms, or talking to Google as if it were a person, computers often can’t process that request because they don’t understand the requestor have never seen it before. RankBrain manages to generalize the sentence: ‘This sentence looks like something I’ve seen in the past, so I’ll assume that’s just what you wanted to know. It’s like a person talking to you in a crowded bar – they can’t hear everything you say, but they can still guess what you mean and continue the conversation with you.”
In that interview, Greg further stated that, shortly after its release, RankBrain became the third most important ranking factor. Since it has such an impact, it’s important to understand exactly how this algorithm works and what changes it can bring to users and SEO experts.
How does RankBrain work?
As I said earlier, RankBrain’s main objective is to deliver more relevant results, interpreting the full meaning of the sentence instead of focusing on individual words. This algorithm is able to handle complex long-tailed search requests well, understand how they are connected to specific topics, and provide relevant results.
In a nutshell, RankBrain identifies patterns in different search requests (even those that seem completely unrelated) and finds similarities between them. This allows Google’s search engine to understand a phrase that has never been seen before by simply matching it with phrases already known to the robot.
As a machine learning system, RankBrain is constantly self-learning, probably paying attention to metrics (eg bounce rate or time spent on the page). That is, if a user believes that the results presented are not relevant, the next time the algorithm will show other results for that search.
What does this mean if you are a Google user?
- You will be able to find information about a thing, concept or fact without using that specific word in your search (the example used by Bloomberg is “What is the name of the consumer at the top of the food chain”);
- You will get more relevant results for search requests that are ambiguous or that have multiple meanings (eg “Apple” – the brand name and “apple” the word for “apple” in English);
- If you type in a search request that Google has never seen before, it will be correctly interpreted and compared to known requests.
Google does not share the exact algorithms it uses, however, we know that its working principles are similar to those of the word2vec tool.
What is word2vec?
Word2vec is an open source toolkit that uses body text to calculate the distance between words and produce vector representations of words and phrases.
This helps to understand the relationship between words based on the distance between them in texts. Words with similar meanings are close together (in vector space). Chris Moody describes a test aimed at finding the vectors closest to the vector of the word “vacation”.
To learn more about word2vec and its operating principles, check out A Beginner’s Guide to word2vec, written by Distilled. If you are interested in more technical details, read the Vector Representation of Words tutorial, written by TensorFlow.
SEMrush experiment on RankBrain
We wanted to gain a deeper understanding of how RankBrain works, so we did an experiment. We try to build connections between words using the algorithm word2vec and data SEMRush (texts obtained from the Brand Monitoring Tool ). For clearer results, we process only text bodies related to Digital Marketing and SEO.
In the end, we got a tool that can be used to introduce any word from the Digital Marketing branch and receive a list of words that are more related to the initial word. They are not exactly “synonyms” (ie, words that have similar meanings) or “related words” (words that bring similar search results on search engines). It is something completely different – these are words that appear more in the texts together with the introduced word.
With our experiment, we try to understand how Google “thinks” and what words it considers related to the keywords we are targeting. The results were very interesting and at the same time were nothing close to what was expected.
To start with, we did a survey among Digital Marketing experts, asking them to give three associations to some words and then we compared the results with those obtained with the help of the tool.
How to optimize a page for RankBrain?
According to official Google representatives, there is a way to optimize for RankBrain. What’s more – RankBrain will not have a drastic effect on search results, as the main purpose of the algorithm is to deal with searches for which data is missing.
Another point – as Gary Illyes, search analyst at Webmasters said on Twitter, “RankBrain also has no influence on crawling or indexing”.
However, it would be unwise to simply ignore the existence of such a powerful algorithm in your SEO activities. So what conclusions can we draw about how the RankBrain algorithm works?
1. Expand your keyword list beyond synonyms and related words.
No more creating pages and content optimized for one keyword or keyword phrase. For maximum effect, try including these elements in your semantic core:
- Your keywords, their variations and related keywords (chosen with the help of your favorite keyword research tool);
- Additional words that most appear in the same context as the targeted keywords.
2. Focus on creating comprehensive content that delivers value to your audience
Write more complete posts, try to talk about all aspects of the chosen topics, answer as many questions as possible. The ultimate goal of Google and RankBrain is to ensure that users get better, more relevant results. If you share the same goal, you’re more likely to succeed.
3. Optimize for people, not search engines
While this advice may seem trite, it is especially valid for machine learning cases. As Neil Patel said in his article :
“It is called machine learning because the machine learns not only from abstract environmental forces, but mainly from the behavior of human beings”.
In other words, there is no point in trying to please search algorithms. Focus on providing a better experience for your users, analyze their behavior and optimize accordingly. If people come to appreciate your content and find it relevant, algorithms will do the same in a natural way.
In conclusion, our attempt to understand one of Google’s most mysterious algorithms has demonstrated that it is nearly impossible to intuitively predict how Google thinks. To make matters worse, being a machine learning system, it is constantly evolving. And at the same time, Google’s algorithms are being updated and optimized.
All you can do to keep up with these constant changes is to make sure you’re always competent and committed in everything you’re doing or writing. That way you will be successful whatever new algorithm or update appears on the market.