On October 21, 2019, Google began to roll out BERT in its search system for English queries. It also included featured snippets, a useful format for users who need concise and direct answers. BERT will affect 10% of searches (source: Search Engine land), which means that 1 out of 10 queries will have an effect in terms of ranking results. In the near future, Google hopes to release BERT models in other languages.
BERT is a Natural Language Processing (NLP) model that stands for Bidirectional Encoder Representations. According to a Google blog, BERT was developed to close the gap in improving context-free language representations on the web. BERT is pioneering unsupervised and deeply bidirectional language representation (source: Google AI Blog). In simple terms, it helps Google understand human language better to serve more relevant results.
What Can BERT do?
BERT uses the context of all the words and it is bi-directional.
BERT utilizes the context and relations of all the words in a sentence. It helps better understand the nuances of words. This means that the algorithm can figure out the entire context of a single word by analysing the words that come before and after it.
Making question answering systems got easier.
With BERT, anyone can do their question answering system in 30 minutes or so on a single Cloud Tensor Processing Unit (TPU).
BERT is a rocket boost for NLP.
BERT is able to outperform 11 of the most common NLP tasks. It is an unsupervised NLP model, meaning you can feed it data and it is able to create clusters or associations among the data points.
Why is BERT important?
It is valuable for highly specific content.
By using BERT, Google can understand the full gist of a query, so a website’s content should still be specific and relevant. Websites who exactly answer the searcher’s questions are more likely to rank higher in Google search. BERT is all about answering a searcher’s question as quick as possible. If for example the owner of a local business is searching for the phrase Fresno SEO in Google since this is the location of his business, than he will find from now on at the very top of Google’s rankings websites which best answers his search term (content should be specific about local SEO services offered within the city Fresno, etc.)
It focuses on informational Google search phrases.
BERT focuses on informational search phrases, which means that its main impact is on top-of-the-funnel keywords. Generally, people do three levels of queries, which are likened to a funnel: informational, navigational, and transactional. Most funnel keywords are informational search phrases. (Source: Neil Patel)
BERT affects featured snippets in Google rankings.
BERT is used for featured snippets in different languages. It can positively impact featured snippet results by deeply understanding the meaning of the query, yielding more relevant results.
BERT works best in more complex queries.
Content optimised for BERT is highly-specific and uses long-tail keywords as well. That is why it is important to analyse competitors, especially those that have a higher rank because their content will most likely have more matching search intent.
Probably the best takeaway that we can get from Google’s BERT update is that the search engine is now getting closer and closer to understanding the language of humans. As for rankings, Google will now be able to show results that are a better fit for the query.