Last week Google revealed it has been using RankBrain, a machine learning process, to help return results for a large number of search queries. Google began a slow, quiet rollout of the update in early 2015, and RankBrain has been completely live and in use for searches globally for the past several months.
With the addition of RankBrain to its algorithm, Google is attempting to better predict what information people are searching for by using past queries to learn user intent. This will help Google return higher-quality results for new complex queries. In an interview with Bloomberg, senior research scientist, Greg Corrado said RankBrain is now processing a “very large fraction” of search queries.
From strings to things
In the early days of search, Google was looking for specific strings — or combinations of letters — on web pages. For example, a search for “estate planning” would return only pages containing that specific string of letters. This reliance on strings made keyword stuffing a popular SEO tactic. The more often you could fit your keywords onto a page the better.
Google, of course, matured and got more proficient at recognizing similar words through a method called stemming. With stemming, a search for “lawyer” would also return pages containing the word “lawyers,” since Google could recognize those to be variations of the same term. It also began to learn synonyms, so that a search for "lawyer" might return pages containing the words “attorney” or “attorneys.”
Moving toward intelligent search
With the release of Hummingbird, Google began to move away from looking for specific letter combinations and toward looking for things, or entities. An entity is a person, place or object — a concrete data point. This allowed Google to start looking for relationships between search queries and other entities. Post-Hummingbird, a search for “estate planning” might return pages containing terms Google has decided are related to estate planning, like "trusts" or "advance directive."
Google is very good at establishing relationships with known entities. For example, when you search for “Ginsburg,” Google assumes you mean Supreme Court Justice, Ruth Bader Ginsburg. It also suggests that you may be misspelling the last name of poet, Allen Ginsberg.
With Knowledge Graph, Google can draw connections between data points. Because of this you can find information on an entity even if you don't know they exact term you are looking for. For example, Google can also return results for Ruth Bader Ginsburg when you search for “Jane Ginsburg mother.”
Meeting the unknown
Every day, however, hundreds of millions of searches are performed for terms Google has never before encountered. In these instances, Google is less proficient at making correlations that the human mind would naturally make. It tends to throw out a kitchen sink of results — pages it is guessing could be related to the original search query. Some are relevant. Some may not be.
How can results for these complex, unknown queries become more accurate?
As smart as Google's algorithm has become, its intelligence still largely depends on an actual human somewhere entering some sort of data. RankBrain is Google's attempt to add a much greater degree of automation to the process.
Search Engine Land estimates Google handles more than 450 million queries a day that contain a combination of words it has never seen. Many of these are likely multi-word, long-tail searches. RankBrain is intended to help handle these unknowns by learning to recognize patterns and connections and use them interpret new search queries more effectively.
As the algorithm becomes smarter, it will be able to better understand the millions of new complex queries it sees every day and determine how (and whether) those searches are associated with known topics. This will help it better predict user intent and understand which pages will be most helpful to the searcher.
RankBrain does not replace Hummingbird. Rather, it is a part of the overall Hummingbird algorithm. According to Google, RankBrain has quickly become a major piece of that algorithm. In his interview with Bloomberg, Corrado said:
“In the few months it has been deployed, RankBrain has become the third-most important signal contributing to the result of a search query.”
A ranking signal is any factor Google uses to determine how to rank web pages in results. Ranking signals are usually associated with page content. Words, titles and mobile-friendliness, for example, are all ranking signals.
Now, Google is referring to a piece of its algorithm as a ranking signal. It is still unclear exactly what this means. Will RankBrain produce a score, similar to Google's mobile and user-friendliness scores, that will contribute to a page's ranking? Is it assessing quality by analyzing related terms? Google may answer these questions in time, but for now it is remaining silent about the details.
Does this change search marketing?
If you are already using white hat search marketing techniques that fall within Google's guidelines, not a lot will change. Google's algorithm is attempting to learn to think more like an actual human being so that it can provide better results and a better user experience.
Google's foray into machine learning likely indicates that Google is anticipating further growth in voice search and mobile search. Voice queries are by their nature usually more complex. It also confirms Google's interest in micromoments - or specific periods of time (and place) for which searchers want a particular set of results for a given query.
None of this is new or revolutionary, and Google has been moving in this direction for many years. You still need to focus on providing helpful, high-quality content, easy to navigate pages and a mobile-friendly experience focused on the visitor. You also need to continue to accumulate social proof, especially in the form of reviews. As Google's machines continue to learn, the algorithm should reward those who have been doing SEO the right way all along.