A (truly) major update of the algorithm
Google rarely communicates about new algorithms – to the great displeasure of specialists. Several times a year, it informs the SEO community when a larger update than the others is deployed – but Google rarely focuses on the scope of its updates. Except for this time, Google did it. They did so because they consider this update to be “the greatest improvement in search in 5 years, and one of the greatest achievements in the history of search”.
Better understanding the natural language of Internet users
The objective of the update deployed by Google is to better understand the natural language of Internet users. The engine notes that most Internet users prefer to use keywords when searching on Google, rather than naturally asking their questions. They do this because Google often offers more relevant results with keywords rather than more natural queries. It is once again thanks to the machine learning that Google will “significantly improve the understanding of requests” and better understand the requests of Internet users.
BERT, to determine the precise meaning of complex queries
Since last year, Google has been developing a neural network-based technique for natural language processing called BERT – for “Bidirectional Encoder Representations from Transformers”. The goal is to better understand the relationships between words within an entire sentence, rather than to deal with expressions and words one by one. Google tries to determine the precise meaning of your requests, sometimes very precise, where the use of certain link words, in particular, is not insignificant.
The impact of BERT on research results
In concrete terms, BERT will impact search results and featured snippets. According to Google estimates, in the United States, 10% of English searches are concerned. The engine will deploy this algorithm update in other countries and languages over time. Google insists on the international scope of its update, stating that it can use data in English to improve its model in other languages. It, therefore, does not need a foreign language dataset to apply these changes to the results in said foreign language. The longest queries, those made using many words with prepositions, those related to natural language, are concerned.
Google, before/after BERT
Google offers several examples that show the changes made by BERT.
2019 brazil traveler to usa need a visa: the word “to” is found, Google now understands that the request concerns a Brazilian who wants to go to the USA. Previously, Google could return results describing the formalities required to travel to Brazil for US citizens.
Do estheticians stand a lot at work: Google matched the word stand in the query to the keyword stand-alone in search results. Thanks to BERT, Google now understands that Internet users wonder about the physical efforts associated with this profession and present more relevant results.
Can you get medicine for someone pharmacy: Google now understands the notion of “for someone” and adapts the results accordingly.
Maths practice books for adults: in the same spirit, Google better interprets the context and target identified in a query.
Parking on a hill with no curb: Google now understands the words that introduce a negation. Previously, we could obtain results that corresponded to the opposite of our need because the search engine focused on keywords by omitting certain words representing a negation.
The impact of BERT should, as such, be strong, particularly on requests using natural language. The fact that the relevance of the associated results is improving, as well as the increase in voice requests, should also reinforce the consequences of this innovation on the results obtained on Google. For web professionals, it is advisable to continue to design and write your content with the intentions of Internet users in mind.
By integrating this type of update into Google’s algorithms, artificial techniques should become less effective in favour of well-thought-out, well-written content that meets the needs of Internet users.