tel: 01895 619 900   email:

Rankbrain: Google AI to filter search results

Google rankbrain

AI (Artificial Intelligence) has been interpreting a large percentage of your Google searches for the past couple of months. Nicknamed Rankbrain, this system embeds large amounts of written language into mathematical entities that are understood by the computer. If it comes across a word it isn’t familiar with, the system is intelligent enough to make a guess and use that to filter the results. To put the system to the test, Rankbrain was pitted against some of the top Google search engineers to guess the rankings of some pages. The engineers were correct 70% of the time while Rankbrain outdid them at 80%.

Not long ago it would have seemed an idiotic notion that companies would entrust their valuable businesses to an AI but that is exactly how it is now trending. Last week Greg Corrado, in an interview with Bloomberg, explained the whole concept of Rankbrain and how it is being developed. 

The post-Hummingbird Era

Rankbrain is the first big news since the launch of Hummingbird which added Semantic search into its search engine. This meant that it didn’t simply look up each of the words in the search but how each of those words made up the entirety of the search query. At that time, Google was also promising to understand and process natural language (or NLP).

From ‘strings’ to ‘things’: the Semantic search

The Semantic search was two-pronged:
First off, it was built to merge search behaviour between desktop and mobile (mobile searches had and have been driven by virtual assistants such as SIRI, Cortana and Google Now);
Secondly, it sought to improve search accuracy by understanding the searcher intent finding not only the keywords but also the contextual meaning of terms. These searches would be resultant from methodologies such as keyword-to-concept mapping, graph patterns and fuzzy logic.

The Knowledge Graph

With the release of the Knowledge Graph in 2012, Google started moving from 'strings' to 'things', but this proved to be much more accurate with data-based information rather than complex query sets. It provided structured and detailed information regarding a search as well as a list if external links.

Is it an algorithm?

Although the info released by Google is pretty unclear, Rankbrain doesn’t appear to be an entirely brand new algorithm, it's just a part of Google’s overall search algorithm. We do know that, when offline, Rankbrain is fed with batches of historic searches from which it learns by matching the search results, these are then verified by the Google engineers and sent live.

For many, its seen as a new SEO signal/ranking factor.

SEO signal

Ranking signals are the characteristics of a website that search engine algorithms consider when valuing its rankings. There are hundreds of ranking factors - see here and here

The third most important SEO signal

According to the Bloomberg article, Rankbrain is only the third-most important signal but, as they say in ‘Search Engine Land’, it is probably being hyped up to give Google some much needed PR regarding what it considers to be its AI breakthrough.

In fact, apparently its main use is to look at the search detail entered and to interpret that in different ways in order to point toward pages that may relate to the search but not actually use the exact wordage the searcher has used.

There has been a gradual roll-out of Rankbrain since early 2015 but it is now fully live and global

Machine learning and AI: is it sci-fi or SEM?

Greg Corrado in his interview used phrases such as "learning system" and "AI", but it’s also fair to say that Panda is also a machine learning-based algorithm that has the ability to learn through repetition what does or does not constitute a “quality website”.

As we understand it, Rankbrain is designed to interpret queries and translate them in order to find the best results for the searcher. When it doesn't know a meaning or it finds words or sentences unfamiliar, it will be able to make a guess by finding something that has a similar meaning.

Understanding natural language & queries

Like Hummingbird, Rankbrain generalises and rewrites queries, trying to match the intent behind them and in order to achieve that, it uses vectors and their relations with queries.

At the moment, we are curious to learn more on how it actually is able to create the vectors from the content it analyzes, as it will have to go beyond the same old "back links, anchor text and surrounding text".

Probably Thomas Strohmann's patent filed by Google is behind the Rankbrain scene, as Greg Corrado mentioned Strohmann’s name as one of the AI experts.

What's next?

It’s most probable that in the next year nothing will change or at least, as users, we won't see any difference and, as web marketers and SEOs, we won't need to change our current strategies and techniques.

Rankbrain targets non-understandable queries and those using colloquialisms therefore, although tomorrow Google might publish Rankbrain guidelines, we won't necessarily be able to understand them.

Nowadays, it's a good practice to produce your content to answer questions and not just to target exact keywords and optimise them by using semantic SEO practices.

In other words, if you are doing white hat SEO and semantic optimisations, you won't have any issues, only benefits.

Some sources to learn more on Rankbrain

Pin it