The BERT Update – Why Google understands us even better now

With the latest update, Google has improved its search algorithm yet again. The technology used is called BERT and is intended to ensure that the search engine understands entire sentences better in the future.

To start with, what do ELMo and BERT have in common?

They are not only the names of two characters from the well-known children’s series, but also the names of two of Google’s speech recognition techniques. Both techniques are used to analyze and mimic human speech with computers – the field of Natural Language Processing (NLP).

As Google announced on its Google Blog on October 15, 2019, BERT will henceforth be included in the search algorithm through an update.

According to the company, this is the most significant evolution of the search algorithm since the introduction of the Knowledge Graph five years ago. Back then, the algorithm started “thinking” in terms of entities – now it can understand complex relationships between them. The following image shows how accurately the search engine can now understand the user’s intent.

Previously, search queries were processed word by word, entities were recognized and related to each other. Conjunctions and prepositions were not taken into account. However, this often failed to capture the deeper meaning of a sentence.

With the BERT update, the search algorithm now places words in the context of the preceding and following words and takes conjunctions and prepositions into account. In this way, it recognizes contexts of meaning that result from a particular sequence of words.

So far, however, the BERT update is only available for US Google search. But German Google users are already benefiting from an improved search algorithm.

Featured Snippets are already being played out using BERT technology in 24 countries, including Germany. Featured snippets are located above the actual search results and provide concise answers to user questions without the need to visit a website.

With BERT, Google can now understand even better formulated questions, so that Featured Snippets can be determined in a much more targeted manner.

It’s short for a technique called Bidirectional Encoding Representations from Transformers and is used by Google for NLP pre-training.

BERT was developed using Tensorflow and is based on a subfield of machine learning called neural networks. These are comparable to the neurons of the human brain, which can independently learn from experience and transfer their knowledge. So you help Google to understand your content by using structured data on your website. Since building neural networks is very computationally intensive, Google had to upgrade its hardware with special TPUs, powerful tensor processing units, for the update.

Google released BERT’s technology under the open source license back in November 2018, but it wasn’t included in the algorithm until a year later. 

According to Google, with BERT, the search algorithm can better understand the user’s intent for 10% of all search queries and thus deliver better search results. This affects both long-tail and nested search queries, as well as formulated questions.

As with every update from Google, the question arises as to how websites can best be optimized for this. Danny Sullivan of Google already clarified in a tweet that optimizations are not necessary.

SEOs therefore do not need to worry about possible optimization measures. However, the update should be seen as an incentive to write natural texts for humans instead of machines. You already get this by following these three tips:

Google itself announced in 2016 that 20% of all mobile searches were already made via voice. Even though there have been no new figures on this since then, it can still be assumed that voice search will gradually replace text search. Likewise, the use of smart speakers such as Google Home or Amazon Alexa will continue to increase.

The fact that Google can now understand entire sentences will not only be an advantage in the competition with other search engines, but also a decisive factor for Google’s top position on the market.