«People's curiosity is infinite. Every day we witness billions of searches, and 15 percent of these queries are new: we have never seen them before. We are therefore building systems to provide results for those searches that we cannot anticipate.».

This is how Pandu Nayak, Vice President for Search at the Mountain View company, comments on the launch of BERT, on official blog by Google. An algorithm update that feels like a revolution. Nayak himself explains that it is "of the biggest leap forward in the last five years and one of the most important in the history of Search” on Google.

Google Bert: what it is and how it works

BERT is a system that leverages NLP, Natural Language Processing: the computational processing of natural language. What is meant for natural language? It is the language we actually use every day: it is not the set of rules and vocabulary, but rather refers to what we say to each other daily and what we write, for example, in chats. Compared to formal language, natural language is much more complex for a machine to understand and process because it contains many nuances, ambiguities, and often cannot be interpreted without context.

Here: Google wants to become proficient at understanding natural language, which is what we often resort to when conducting a search (consider the exponential increase in voice searches in recent years). BERT will therefore help the algorithm better understand our search keys, commonly referred to in jargon query. Explained in simple terms: Google should now give more importance to context and entire sentences rather than individual words.

Nayak emphasizes how the new system also leverages the so-called machine learning, machine learning by a "machine": in practice, artificial intelligences can (within certain limits) learn on their own by analyzing available data, for example on the Internet.

For the time being, the new algorithm operates at full capacity only with English and today it is already capable of better learning about ten percent of queries in the United States. However, as Nayak further explains, the new update will soon be operational in various locations around the world, with different languages (tests are already underway in this regard).

How BERT was born

In 2018, Google launched an interesting experiment called Bidirectional Encoder Representations from Transformers, from which the acronym BERT. It is a platform open-source, to test the possibilities of neural network techniques applied to natural language. A technology that has allowed anyone to test their own automated response systems.

The true innovation brought by this technology lies in the T of Transformers. These are the logical models that allow for the analysis of words in relation to all others in the sentence, rather than one after the other. 

To effectively implement the new update, Google also had to enhance the hardware, not just the software. BERT will indeed use the TPU (Tensor Processing Unit), an artificial intelligence accelerator consisting of an ASIC circuit (which Google itself developed). The system is essential for applications in the neural network sector. The TPU used for BERT is also in cloud.

What changes with Google Bert

Let's now try to understand what actually changes with the advent of BERT. Google has already provided some examples to make everything clearer. Among the most interesting, there is this one about Taylor Swift:

Searching the query on the search engine "How old was Taylor Swift when Kanye went on stage?", previously Google would return a video of Kanye West at the MTV Video Music Awards. After the update, the first result is a BBC article that exactly answers the question.

A second interesting example comes from Search Engine Journal, an online newspaper specialized in SEO and SEM optimization. The author, Roger Montti, illustrates a small personal "experiment":

The query in this case is untranslatable, "how to catch a cow fishing?". This is a colloquial expression in the field of fishing, limited to the New England area. Until recently, the word "cow" (mucca) misled Google. While the cow fishing it is indeed a fishing technique, the search engine returned a series of results related to farms.

The experiment was repeated by Montti on October 25, after the release of BERT: in this case, Google actually “understood” what the user was searching for.

Here is a clearer explanation of how BERT understands the overall context of each query, rather than individual words. It is essential to understand this because it means that Google will focus more on clarifying the so-called "search intent", that is, the intention or need behind the search. For Google, it will no longer be important to understand phrases word by word, but rather to intuit what drives human curiosity, what is the necessity behind the search.

As can be easily understood, SEO optimization professionals will need to be careful to understand Google's new approach to queries. 

According to initial analyses by experts, the search engine will reward those who use appropriate terminology based on the specific context discussed in the articles. Generic language could be penalized in favor of greater attention to specialist expressions. (as we have seen in the case of the cow fishing). In any case, it will become increasingly important to try to "get inside the user's head": why are they searching for that specific content? What do they need it for? What need does it satisfy? These questions will need to be answered precisely.

It will be increasingly less important to stuff articles with keywords, increasingly it will be useful to understand the search intent. It will be more effective to create content where the correct terminology is used based on the specific context to which it refers, obviously providing answers more to the search intentions rather than relying exclusively on the number of keywords used.

However, the page structure will not become obsolete. After all, clear content that provides simple and immediate answers is also well organized. As further emphasized by Search Engine Journal, the algorithm expert Dawn Andersonwe must emphasize the importance of using clear structures.