Google RankBrain

Google RankBrain is the machine learning system Google rolled out in 2015 to help interpret search queries - particularly novel queries the system has never seen before - and rank results more relevantly. RankBrain converts queries and content into mathematical vectors and matches them by semantic similarity rather than just keyword overlap.

The first major machine-learning component embedded in Google’s ranking stack. Set the precedent for the heavily ML-driven systems that followed (BERT in 2019, MUM in 2021, the various AI-powered ranking signals shipped since). Established that exact keyword matching was no longer the dominant ranking lever.

What RankBrain changed about ranking

Three shifts that mattered immediately:

Synonyms and concepts became first-class. A page about “how to grow my email list” started ranking for queries about “increase newsletter subscribers” without the exact phrase appearing on the page. Semantic understanding replaced literal matching for the long-tail of queries.

Novel queries got better answers. 15% of daily queries are queries Google has never seen before. RankBrain made a meaningful difference for these because it could pattern-match to similar past queries even when literal keywords didn’t overlap.

User behaviour signals weighted more in ranking. RankBrain incorporated behavioural signals (click-through rate, dwell time, return-to-search rate) into ranking iteration more heavily than the previous system. Pages that satisfied users got promoted; pages users bounced from got suppressed.

What RankBrain didn’t change

Worth being clear about:

Backlinks still matter. RankBrain is a query-interpretation and result-ranking system. Backlink-derived authority signals continued to be a major ranking factor; RankBrain adjusted relevance ranking on top of authority signals.

On-page SEO still works. Title tags, headings, content structure, internal linking - all continued to matter. RankBrain made keyword-stuffing less effective and semantic matching more effective, but the fundamentals of well-structured content didn’t change.

What RankBrain implied for content writers

Three operational rules that emerged:

Write naturally for humans, not for keyword density. Trying to fit “best email marketing software” into a page exactly seven times became unnecessary and counterproductive. Writing about email marketing software comprehensively, in natural language, got the same coverage with better readability.

Cover the topic, not just the keyword. A page targeting “email marketing software” should naturally cover related concepts - pricing, segmentation, deliverability, common alternatives. Topical comprehensiveness signals to RankBrain that the page is a good answer for the broader query intent.

User satisfaction signals are a ranking moat. Pages users stick on, scroll through, and don’t bounce from get rewarded. Pages users bounce from get suppressed even if they technically rank well on other signals. UX is SEO.

An example

A B2B SaaS team had written a 1,200-word article targeting “best CRM for small business” - keyword in the title, repeated 11 times throughout, exact-match anchor text on internal links pointing to it. Ranked #11 for the query in 2017.

The 2018 rewrite (post-RankBrain awareness) was a 2,500-word genuinely useful article covering CRM categories, pricing tiers, integration considerations, common pitfalls, and concrete recommendations for three different small-business profiles. The exact phrase “best CRM for small business” appeared twice - naturally - in the rewrite. The page targeted the topic, not the phrase.

Six months in: ranking lifted to #3. CTR went up because the title matched what searchers found. Time-on-page tripled. Conversion to email signup from the article was nine times higher than the keyword-stuffed version. Same query, better page because it served the searcher rather than the algorithm.

We built Penfriend with RankBrain-style semantic ranking in mind. RankBrain rewards content that matches query intent rather than query keywords; Penfriend generates against intent specified in the brief, not against keyword lists.

Related terms

• Google Algorithm - the broader system RankBrain became part of
• Algorithm - the abstract concept RankBrain instantiated for query interpretation
• Google Panda - an earlier major algorithm update with related quality-driven principles
• Anchor Text - the on-link signal that became less dominant after RankBrain
• Featured Snippet - the result format RankBrain often determines candidate selection for