About Google's new BERT algorithm
Let's talk about what Google BERT is. Without understanding the basic principles of this algorithm, you can't effectively promote and grow a business, whatever its niche.
To explain the concept, we first need to look at neural networks and at what natural language processing is.
A neural network is the foundation of the algorithm. It can recognise patterns. It's what finds images by category. It can also recognise handwriting. To improve performance, developers take large datasets for preliminary training and testing. Google BERT, for example, was trained on Wikipedia.
Natural language processing is one of AI's most important tasks. Put simply, its point is to learn to understand how people communicate. That's why we now get accurate suggestions while typing on our phones and analytics on how certain queries behave. Now let's move on to the algorithm itself.
BERT According to Google, this is a very significant update that will help better understand user behaviour and return more informative and useful links.
Google BERT was introduced in 2018. A little later, the company open-sourced the code. This allowed any other company not only to understand how BERT works but also to use it to train its own language analysis systems. In October of the following year, it was rolled out in search.
Bidirectional Encoder Representations from Transformers (that's what the acronym stands for) is the newest method of natural language processing. It directly affects how passages of pages are ranked, so users get the most relevant answers, because it better understands the meaning of search phrases.
How RankBrain differs from BERT
Not everyone understands the difference between BERT and RankBrain. It might seem they're just two names for the same thing. But no — they are independent, so they can be used at the same time, which also improves the quality of search results.
RankBrain is used to adjust other algorithms. It works by comparing a new query with ones that have been used before.
Based on that data, it adjusts results and helps the search engine understand what the user wants, even when keywords aren't used directly.
Bidirectional Encoder Representations from Transformers works a little differently. Earlier algorithms analysed page content to work out what it was about. BERT is bidirectional. It analyses everything both before and after a word, so it understands context better. It can learn language models taking all words into account, whereas previously analysis was only possible left to right or the other way round. That's why it's so effective.
Google BERT and SEO
If you're already panicking about how to optimise your site so you don't fall behind, hold on. Google itself said these updates would affect a very limited number of pages, so the changes will be barely noticeable. Besides, at first the algorithm was applied only to English-language queries. English-speaking users were the first to feel the benefits — results matching the question as closely as possible. Sites in other languages had nothing to fear for the time being.
Still worried? Then just make your content as lively and useful as possible, not an empty set of worn-out phrases written for a machine. And don't forget the importance of information architecture. Place queries from search visibility data and search suggestions sensibly.
Users will appreciate the algorithm, as search relevance is now much higher. It also improves voice search. That's great news, as more and more people browse the internet on smartphones, where voice commands are indispensable.
As for SEO specialists, it lets them tailor meta tags to CTR without compromising optimisation quality. Previously you had to choose: write them for SEO or for CTR. In short, it's a confident step towards better interaction with people looking for information, and an incentive to fill the internet with quality content.
