RankBrain.

RankBrain is a machine learning component of Google's core ranking algorithm, first confirmed by Google in 2015. It processes search queries—particularly novel or ambiguous ones it hasn't seen before—and maps them to known concepts and relevant results by interpreting the meaning behind the words rather than matching keywords literally.

How RankBrain Works

RankBrain uses word vector representations (embeddings) to infer the probable meaning of unfamiliar queries based on the mathematical relationships between words in Google's training data. A query like 'what's the highest sky' can be interpreted as a question about altitude records, not just matched to pages containing those exact words.

RankBrain and BERT

RankBrain handles query interpretation at the query level; BERT (Bidirectional Encoder Representations from Transformers), introduced in 2019, improved Google's understanding of natural language nuance—particularly prepositions and word order that change meaning (e.g., 'can a Canadian visit the US without a visa' vs 'can a US citizen visit Canada without a visa'). Together they make Google's systems significantly better at understanding intent.