Google's BERT technology: what is it and how to live with it?

For several days now, the main topic of discussion among Internet marketers around the world is Google's announcement of a new search technology BERT (Bidirectional Encoder Representations from Transformers) successful launch.

According to Google search representatives, this is one of their main achievements over the past 5 years.

What is BERT, and can you prepare for it?

BERT is a technology for processing user requests and generating search results, which is based on the use of machine learning. Its purpose is to give users the most accurate and correct answer top ambiguous (wordy, interactive) search queries.

In other words, BERT can help computers understand language a little better than humans.

BERT will affect 1 in 10 of all searches. This is the biggest change to search since Google released RankBrain.

What is special about the new algorithm?

BERT is designed to do more than just pick up search results by searching for words that the user has entered. This technology will "read" the query, like a normal person, and capture the hidden meaning. BERT models can take into account the full context of a query by looking at the words that come before and after it, which is especially useful for understanding the search queries. intent

Now BERT is already working on search in the US in English, and Google representatives are already demonstrating the results of its work.

Comparison of Google search results for the query "2019 brazil traveler to usa need a visa" before and after the implementation of BERT:

Previously: The result was information for US citizens about traveling to Brazil.

Now: The user sees resources with visa information for Brazilians who plan to visit the United States.

The introduction of BERT has led to the fact that the search does not just look for the word "to" that the user enters in the query, but clearly understands its meaning in conjunction with other query words.

In addition to the United States, partial implementation of the Google BERT model is confirmed in two dozen more countries, and in addition to English, there are also successes in Korean, Hindi and Portuguese.

How does the introduction of BERT affect online marketing?

The implementation of BERT models is taking place gradually, by country. The search engine representatives didn't comment when exactly the technology will be fully implemented in search around the world.

But everyone is already interested in questions:

1. How will this affect output?
2. How to prepare for the implementation of BERT?

We answer in order.

For users, this means that the chances of getting the most relevant results, regardless of the type of query, increase. This applies especially to:

• low-frequency queries with a "long tail";
• voice search results;
• queries, when the user himself does not fully understand what they are asking about, and enter not exact queries, but near-topic questions or just a set of words that, in their opinion, are related to their topic of interest.

Preparing for BERT … Preparing does not mean introducing something new, unknown. Website owners and SEOs should continue to pay special attention to:

• Content expertise on websites. Not general phrases on the topic, but a detailed description of the product or service. Another argument is that it’s better to pay once for a good, high-quality text (if you can’t write it yourself) than to look for cheap SEO copywriting options that simply won’t work.
• A detailed response to the user's intent. Not just text corresponding to one target search query (or a group of very close ones), but the most detailed answer to all near-thematic hints. It is important to work not only with frequency requests.
• Optimizations for the widest possible pool of queries. This was important yesterday, as this is not just a task of preparing for BERT, but a mechanism for attracting the widest possible target audience to the site.

Finally, I would like to say: the introduction of BERT is not a threat, but an opportunity that can and should be used to your advantage right now! Don't miss it :-)