What is RankBrain?
RankBrain is an AI system in Google Search that links the words in search queries to concepts and can therefore also understand queries that Google has never seen before. Google has been using RankBrain since 2015, and it was the first deep learning system in Google Search. RankBrain helps find relevant pages even if they do not contain all the words of the search query.
Term profile
| Attribute | Details |
|---|---|
| Part of speech | Proper noun |
| Pronunciation | ˈræŋk breɪn |
| Operator | |
| In use | since early 2015, publicly known since October 2015 |
| Type of system | Machine learning (deep learning) for interpreting search queries |
| Scope | initially new and rare search queries, since 2016 all search queries |
| Status | active system in the official guide to Google Search ranking systems |
| Related terms | Google BERT, MUM, Hummingbird, semantic search, search intent |
How does RankBrain work?
RankBrain translates the words of a search query into vectors of numbers and looks for concepts that lie close together in this numerical space. This allows the system to assign an unfamiliar wording to a familiar meaning. The method is similar to an embedding, in which related terms receive similar vectors.
Google itself explains RankBrain using an example query: “what’s the title of the consumer at the highest level of a food chain”. None of these words is “predator”. However, RankBrain has learned from many pages that a food chain has to do with animals and not with consumers in retail. The system therefore recognizes that an apex predator is meant and returns results on exactly this concept.
The process can be described in 4 steps:
- Taking in the query: RankBrain receives the search query with all its words, even if this combination has never been searched before.
- Translating into vectors: Words and word groups are represented as points in a numerical space in which similar meanings lie close together.
- Assigning a concept: The system looks for the concept that comes closest to the vectors of the query and interprets the query in that sense.
- Weighting results: Search favors pages that match the recognized concept, even if they do not contain the words of the query literally.
Why did Google introduce RankBrain?
Google introduced RankBrain because around 15 percent of daily search queries are new and cannot be interpreted reliably with fixed rules. Before RankBrain, Google engineers maintained synonym lists and rules largely by hand. With queries nobody had anticipated, these rules reached their limits.
The history of RankBrain unfolded in 3 steps:
- Early 2015: Google starts using RankBrain in Search, initially for a large share of new and rare queries.
- October 2015: Bloomberg makes the system public. In that report, Google researcher Greg Corrado calls RankBrain the third most important signal for ranking.
- 2016: Google extends RankBrain to all search queries. In March 2016, Google employee Andrey Lipattsev confirms that content and links are the other two of the three most important signals.
RankBrain marked the start of Google’s transition from word lists to meaning. Google BERT followed on this path in 2019 and the Multitask Unified Model (MUM) in 2021. Together, these systems form semantic search.
Is RankBrain a ranking factor?
RankBrain is not a single ranking factor you can fulfill but a system that interprets the meaning of a search query and thereby influences ranking. Its description as the “third most important signal” in 2015 led to the misconception that RankBrain is a factor like load time or backlinks. Today, Google describes RankBrain as an AI system that understands how words relate to concepts.
The role of user signals is disputed. In 2019, Gary Illyes from Google described RankBrain as a component that uses historical search data to predict what users would most likely click on for an unfamiliar query. This does not establish metrics such as the click-through rate or the dwell time of a single page as a direct switch, however. During the US government’s antitrust case against Google, it also became known in 2023 that Google processes click data mainly in a separate system called Navboost.
Also important: RankBrain does not learn in real time from every search. Google trains the model with collected data and rolls out new versions at intervals.
What is the difference between RankBrain, BERT, and MUM?
RankBrain, BERT, and MUM are three AI systems in Google Search that build on one another and have different tasks.
| System | Since | Task | Put simply |
|---|---|---|---|
| RankBrain | 2015 | Links the words of a query to concepts | What does this query mean? |
| Google BERT | 2019 | Understands words in the context of the whole sentence | How do the words of this query relate to each other? |
| MUM | 2021 | Processes text and images and transfers knowledge between languages | Which information from other languages and formats helps here? |
The systems do not replace each other, and RankBrain and BERT run in parallel. Google lists all three in its guide to ranking systems. According to this guide, Google does not use MUM for general ranking but for individual applications. A classic update such as a Google core update, by contrast, is an adjustment to the overall system and not a model of its own.
How do you optimize for RankBrain?
There is no technical setting for RankBrain, and targeted optimization lies in content that covers a concept completely. Because RankBrain assigns related wordings to the same meaning, writing for individual keyword variants loses value. This leads to 4 rules:
- One page per intent: Separate pages for “SEO agency” and “agency for SEO” compete for the same meaning. The result is often keyword cannibalization.
- Meeting the intent: What matters is what lies behind a query. If you fulfill the search intent, you also fit unusual wordings of the same question.
- Covering the topic in its breadth: A text that picks up the related terms and sub-questions of a topic gives RankBrain the signals for the right concept. On a larger scale, this creates a topic cluster.
- Answering rare questions: RankBrain was built for queries that never existed before. Specific longtail keywords are therefore an opportunity for pages that answer a question precisely.
These rules are at the core of ongoing SEO services: Strategy and content are planned by topics and intents, not by individual search terms.
Frequently asked questions about RankBrain
Does Google still use RankBrain today?
Yes. Google lists RankBrain as an active system in its official guide to Google Search ranking systems and describes it there as the first deep learning system in Search.
Is RankBrain artificial intelligence?
Yes, RankBrain is an AI system based on machine learning. However, it is not a language model that writes texts but a system for interpreting search queries.
Is RankBrain used for every search query?
According to Google in 2016, RankBrain is involved in processing every search query. The system makes the biggest difference with wordings Google does not know yet.
