RankBrain explained: Google's first machine-learning ranking signal

Fibvo editorial · 18 June 2023 · 5 min read

In October 2015, Google confirmed that a machine-learning system called RankBrain had become one of the top three signals in its ranking algorithm, alongside content and links. RankBrain was notable for a specific reason: it was Google's first ranking system built to handle queries the algorithm had never encountered before, which Google itself said accounted for roughly 15% of all daily searches.

The problem RankBrain was built to solve

Every day, a meaningful share of searches are phrased in a way Google has never seen exactly before — an unusual combination of words, a very specific personal question, or an ambiguous phrase that could mean several things. Traditional keyword matching struggles here because there is no historical data showing which pages perform well for that exact phrase.

• Novel, never-before-seen query phrasing.
• Ambiguous queries that could reasonably mean more than one thing.
• Long, conversational questions rather than short keyword fragments.

How RankBrain actually works

RankBrain converts words and phrases into mathematical vectors, then uses those vectors to guess which other, better-understood queries are conceptually similar to a novel one. It then applies the ranking patterns learned from those similar, familiar queries to the new, unfamiliar one — effectively translating an unknown question into the nearest known equivalent Google already understands well.

RankBrain does not read your page differently. It changes which pages get considered relevant for a query Google has never seen phrased quite that way before.

What this means for content strategy

Because RankBrain groups conceptually similar queries together, a single well-written page that thoroughly covers a topic can rank for many query variations it was never explicitly written for, as long as the underlying concept and intent match. This rewards depth and topical completeness over trying to create a separate thin page for every possible phrasing of the same question.

• Cover a topic's genuine range of subtopics and related questions on one strong page rather than splitting it across many thin ones.
• Use natural variation in phrasing throughout the content rather than one rigid keyword repeated verbatim.
• Structure content with clear headings that match the actual questions people ask, as covered in our topic clusters guide.

RankBrain's relationship to BERT and MUM

RankBrain, BERT and MUM are often confused, but they solve related, overlapping problems at different depths. RankBrain groups similar queries; BERT (covered in our BERT explainer) understands the grammatical relationships within a single query more precisely; MUM extends that understanding across languages and formats. Together they represent Google's steady progression toward genuinely understanding language rather than matching strings. Google's own overview of how AI is used in ranking is documented in its how search works explainer.

Frequently asked questions

If you want help auditing whether your content actually covers a topic thoroughly enough to benefit from how RankBrain groups related queries, get in touch with our team.

---

Written by Fibvo — Bengaluru and Erode, India — working with clients in other time zones. Questions, or think we have it wrong? Tell us.