RankBrain

Long definition

Google publicly confirmed RankBrain in a Bloomberg interview on October 26, 2015, with engineer Greg Corrado revealing it had been quietly active for several months and was already "the third-most important ranking signal" — a striking admission since Google rarely discusses signal weighting at all. RankBrain was Google's first deeply machine-learning-based component in the live ranking algorithm, predating BERT by four years.

What RankBrain solves: roughly 15% of queries Google sees each day are entirely novel — strings nobody has ever typed before. Pre-RankBrain, novel queries depended on rigid keyword matching against the index, which often failed. RankBrain represents queries as mathematical vectors and finds the closest known queries — a vector-space embedding approach — so it can apply the ranking knowledge of the closest related queries to the new one.

Architectural notes:

• Embedding-based — words and phrases are mapped to dense vectors so semantic similarity becomes geometric proximity.
• Pre-Transformer era — uses earlier neural network designs predating the 2017 Transformer architecture that BERT and MUM use.
• Trained on historic search behavior — ranking signals derived from how users interact with results.
• Refines query interpretation, not pure ranking — works alongside other signals (PageRank, anchor text, content quality) rather than replacing them.

RankBrain remains active. Google's official 2022 update on its ranking systems lists RankBrain alongside BERT, MUM, neural matching, and the deduplication and helpful content systems. Ranking is not "RankBrain's job" — it's one component in a stack of components that together produce the final ordering.

The practical implication for SEO: RankBrain rewards content that genuinely answers the spectrum of related questions in a topic, not just the literal target keyword. A page that helps Google understand the topic well — clear headings, comprehensive subtopic coverage, internal links to related pages — works with RankBrain rather than against it. There is no specific RankBrain optimization, just better topical authority.

Common misconceptions

• "RankBrain is THE Google algorithm now." No. It's one signal in a complex stack. Google's official ranking systems documentation lists RankBrain among many systems including BERT, MUM, neural matching, and others.
• "BERT replaced RankBrain." They coexist. RankBrain handles novel-query interpretation; BERT handles linguistic comprehension; MUM handles multimodal reasoning. They serve different roles.
• "RankBrain uses click-through rate as a ranking factor." This claim circulates in SEO circles but Google has consistently denied direct CTR ranking. RankBrain learns from query-result patterns in aggregate, not from individual click signals.
• "You can write content optimized for RankBrain." You can write content that demonstrates topical depth, which RankBrain rewards because it's the kind of content that satisfies query intent. There's no RankBrain-specific keyword formula.