RankActive Blog

AI Revolution and SEO: How to Optimize for RankBrain in 2018

26 December 2017 George Svash Leave a comment SEO topics

User experience is now more important than ever before. There’s nothing like unwrapping the cutting edge iPhone X or MacBook Pro. No wonder the same applies to websites. Imagine you could get the answer to any question instantly. Imagine that you open up the first result in SERP and see well-written content that perfectly fulfills what you’re searching for. Amazing, isn’t it?

Infographic

The future that seemed to be so far away, has arrived. From rich snippets to the knowledge graph, from Penguin to Hummingbird, Google continues on its path of bettering search results and evolving search experience. We are about to undergo another fundamental improvement, which holds the ability to radically change the search landscape and, quite frankly, disrupt the SEO industry.

As you might have guessed, in this article we’ll discuss RankBrain, machine learning AI system that helps Google to deliver deeply personalized search experience.

The story behind RankBrain

The Internet as it is today contains an incredible amount of information. With Google indexing about 30 trillion web pages, Amazon selling over 372 million products and YouTube processing 300 hours of uploaded video every minute, it becomes harder and harder to find the information that we’re looking for. What’s more, long and unsuccessful searches negatively impact conversions and, therefore, profits of businesses.

There are two ways to structure search results and make them relevant to the query. The first one is to provide different results composition for different areas of research. If you look at metasearch engines like Yippy, you’ll find a variety of categories to choose from. You get diverse results, and it’s up to you to decide which category is more relevant to your query.

However, such approach didn’t gain much popularity, and there are only a few search engines that categorize results this way.

The second way to deliver relevant results is by ranking them higher. The idea is that user can avoid wrong results because they are displayed below the right ones. This approach is used by the most popular search engines, including Google and Bing.

Although it may not seem that sophisticated – you can just compare words in the sites’ content against the words in the particular query – the relevance value of each result will be different for each individual user. But what if two users type in the same query when looking for different info?

Let’s turn to the classic example of “Java” query. A user is searching for Java, but how do we know whether he’s looking for a cup of Indonesian coffee, information about the programming language, or maybe she’s planning a trip to Java islands?

It seems like Google has found a solution. They provide results based not only on the short-term and within-session search history but also following long-term and across-session behavior.

Google makes it possible to display three different results pages for exactly the same query.

How did they manage to get the level of personalization that high? Well, even though RankBrain may not be an ultimate answer here, it’s certainly a big part of the solution.

What is RankBrain?

As we’ve already noted in one of the previous paragraphs, RankBrain is a machine learning artificial intelligence system. It’s used by Google for deciphering the semantic meaning of search queries to provide the deeply personalized set of search results.

Google confirmed the RankBrain’s existence on the 26th of October, 2015. In the famous interview for Bloomberg, Greg Corrado (senior research scientist at Google) stated that RankBrain is responsible for as much as 15% of all online searches. Nevertheless, in 2016 Jeff Dean, a Google Senior Fellow in the Google RankBrain team, stated that RankBrain is “involved in every query.”

We don’t know, for how long they’ve been using AI systems in the core of the algorithm, but it’s evident that RankBrain has significantly changed the search landscape. In the pre-RankBrain times, Google would apply the old algorithm – based mostly on keywords and links – to determine the relevance of results to a query. Basically, there was a big team of developers, who would create a mathematical algorithm responsible for determining search rankings. That algorithm would be a constant until an update was made. However, these days are long gone, along with keyword-stuffing and link building practices.

With the launch of RankBrain, Google has made the radical shift from matching searches to determining the real intent of a user. In this case, supposedly, the query goes through some interpretation model, that applies different parameters, e.g., location, search behavior (including long-term and across session behavior), language and words of the query in order to determine the user intent.

“User intent is the identification and categorization of what a user online intended or wanted when they typed their search terms into an online web search engine for the purpose of search engine optimization or conversion rate optimization.”
Jansen, Jim. Understanding Sponsored Search: Core Elements of Keyword Advertising.

Here’s what Google’s RankBrain does in general terms:

    • Interprets the query
    • Determines user intent
    • Selects results and gives out a personalized SERP

Lets use the example of “rent a car uk” to see, how Google handles the interpretation of queries.

Go to the Google Keyword Planner and try to request search volume data and trends.

Although you typed in “rent a car uk,” you would get data and trends for a different keyword “car rental uk.” That happens because, from the Google’s point of view, user intent shall be the same for both terms.

In addition to that, you might have noticed the extraordinarily high search volume for a given keyword. The reason for that is Google showing the cumulative data for a group of semantically similar keywords. So, broad match keywords are being lumped together alongside with the data on their search volume and trends.

So, RankBrain knows that both “rent a car uk” and “car rental uk” fall within the sole search intent. Therefore, both searches are essentially asking for the same thing: “Where can I rent a car in the United Kingdom?”

Another example of Google’s RankBrain in action can be found at Google News.

As you see, the main news is about the stormy weather in New England. The NBC’s headline is displayed on the top of it, but there are also stories from CBC, Pittsburgh Post-Gazette and Seattle PI. How did Google determine which stories are relevant to the topic? Through the help of RankBrain Google determines similar patterns, so it can recognize objects with similar properties even without training examples. Thus, we may conclude that RankBrain is an unsupervised machine learning system that uses clustering to provide relevant search results.

Parameters that come into play

It’s hard to say what parameters RankBrain uses to personalize a results page. As noted above, it’s the unsupervised machine learning system, which distinguishes RankBrain from the wide-spread Learning to Rank architecture.

However, we’ve tried to determine a few factors that, in our opinion, are likely to have an impact on personalized rankings:

      • User. It’s no secret that Google and other search engines are tracking users over time via cookies, IP addresses and logged in accounts. This data helps to predict what results we’d like to see in SERP.
      • Query. Google is reportedly using RankBrain to deal mostly with unique searches, i.e., searches it has never seen before. So, it’s likely that RankBrain is processing only unique queries. Nevertheless, we can’t say whether Google is going to keep it this way in the future.
      • URL. One of the parameters that Google takes into an account is the website authority. Sites using HTTPS protocol get higher positions than those using HTTP. Talking about personalization, we suppose that RankBrain is dealing only with the first page of SERP, since a vast majority of users never go beyond the first page.
      • Click. After user issues a query, they can click on the SERP results. Assuming that clicking on the result is the indication of user interest, Google gives higher rankings to websites that draw the highest interest. We suppose that the last click of the session is the most crucial because a user is not going to terminate the search before she is not satisfied.
      • Click-through rate. If a user doesn’t interact with the website she visited, it means that such resource does not represent any value for a particular user. In that event, rankings of such webpage will sink in the personalized SERP. By the contrast, if a user clicked on the result multiple times within the session, or across different sessions, such website would presumably get higher rankings.
      • Dwell time. The amount of time between a click and the next click (or the following query) is the indicator that Google is supposedly using to measure the potential utility of a page for individual users. The more time user spends on the page, the greater utility it represents.
      • Relevance. Proceeding from CTR and dwell time, Google determines the relevance of results to individual users. The optimal dwell time and CTR – and therefore the indication of relevance – may be different for every query.

What’s all this mean to the SEO community?

RankBrain may seem quite sophisticated, but it’s just another novelty that aims to improve search result. RankBrain only measures how users interact with websites, and ranks them accordingly. Moreover, it’s reportedly rated as only the third most useful ranking factor. It’s not that dramatic so far.

Things will get more complicated when RankBrain becomes the #1 ranking factor. With the development of second generation Tensor Processing Units and other advances in computing performance, there is nothing that would constrain Google from the broader application of Machine Learning technologies. We can’t say if it’ll happen in 2018 or subsequent years, but we definitely should start preparing for it today.

Optimizing for RankBrain

The best advice would be to think beyond keywords. No, you don’t have to stop targeting terms that people use when searching for your services. Just look at them from a different angle. In addition to that, it’s right the time to modify your content strategy.

Here are a few tips that may stand you in a good stead:

Stop using “one keyword per page” technique. Given that RankBrain is actually a part of the Hummingbird algorithm, it’s better to have one page targeting all the keywords rather than multiple pages with one-two keywords on them. You may also try to use LSI (Latent Semantic Indexing) phrases to represent different ways to say the same thing in the content.

Focus on engaging content. To determine the relevance of content, RankBrain might be using Dwell Time and Click-through rates. So, engaging and useful content will become even more important.

Keep your content fresh. RankBrain is looking for fresh content that matches the search intent. Review the existing content to see what updates you can make. You may also consider creating new content more frequently.

What’s your take on RankBrain? Will it disrupt SEO in 2018? Tell us what you think in the comments below.

Tags: ,

Like this article? There’s more where that came from.

mautic is open source marketing automation