Google announced BERT on October 25, 2019 in the blog post "Understanding searches better than ever before" by search VP Pandu Nayak, describing it as "one of the biggest leaps forward in the history of Search". BERT (Bidirectional Encoder Representations from Transformers) is a neural network technique for natural language processing that considers the full context of a word by looking at the words that come before and after it. Google said "BERT will help Search better understand one in 10 searches in the U.S. in English"; Search Engine Land reported it had already begun rolling out the week of October 21.
At launch, BERT applied to ranking for US English queries and to featured snippets in the two dozen countries where that feature was available, with Google citing significant improvements in languages like Korean, Hindi and Portuguese. Google's example query was "2019 brazil traveler to usa need a visa": before BERT, Search missed that the word "to" mattered and returned results about US citizens traveling to Brazil; with BERT it grasped the direction of travel. The models run on Google's Cloud TPUs.
BERT was not a penalty and there was nothing to fix: Google said sites cannot optimize for BERT any more than they could for RankBrain, and the advice remained to write content for users. On December 9, 2019, Google expanded BERT to more than 70 languages worldwide; it had previously applied only to featured snippets in languages other than English.