The effects of machine learning on rankings and SEO

For a long time search engines relied on static ranking factors. Those webmasters and SEOs who knew what to pay attention for were able to reach the best positions on Google’s SERPs. This has changed recently and will be changing in the future: The increasing usage of machine learning techniques leads to both dynamic ranking criteria and – as confusing as it may sound – a greater influence of human signals.

The rise of machine learning

Machine learning is nothing new. Its roots go back to the 50s of the last century. What is new today is the massively grown amount of resources in the data centers of Google, Facebook and other big companies that allow the application of machine learning algorithms in real time. One example for such a powerful cluster is the Google Brain or “Google’s Deep Learing Project”.  In a test, the cluster, consisting of more than 16.000 microprocessors, was able – using the method of unsupervised learning – to learn from a number of 10 million YouTube images what a cat is and what a cat normally does. The most remarkable fact about this is: Google Brain had never been told the concept of a cat before.

One big step Google took towards leveraging machine learning techniques was the implementation of the Hummingbird algorithm in 2013. With Hummingbird Google was able to understand even complex search requests and conversational speech.

And finally last year, RankBrain was introduced – according to Google sources the third-most important ranking factor these days – even if its true importance is not completely clear. The RankBrain algorithm gives Google the ability to interpret even search requests that never occurred before. And there are many of them: According to estimations about 15 percent of the daily search requests Google has to deal with are totally new and never occurred before. RankBrain uses Google’s knowledge base to understand these search requests using a network of millions of different entities and connections between these entities. Consequentially, RankBrain’s impact is greatest for these new, first-time search requests.

The machine learning-based ranking circle

Machine learning and dynamic ranking factors can be seen as an iterative cycle. As mentioned before, user signals play an important role in this cycle – as well as actions taken by webmasters and SEOs:

As a first step Google looks at the most reputable and popular websites – let’s call them seed pages. These websites not only have lots of high quality and relevant backlinks. Additionally the user signals for these websites show a low bounce rate, a great ratio of returning visitors and a long dwell time.

The characteristics of these seed pages are taken into account by the search engines’ machine learning algorithms. The algorithms adjust the ranking factors in a dynamic and continuous way. In return, these adjusted ranking factors are applied to rate other websites. The websites are changed and updated by webmasters and SEOs to satisfy the ranking factors. This has two effects: First, it influences the websites’ rankings on the SERPs and second, it influences the user signals – the starting point for a new cycle.

What about other ranking factors?

This doesn’t mean that rankings in the future will be based on machine learning algorithms alone. In a recent Webmaster Hangout John Mueller from Google told us that relying on a single ranking algorithm would make it hard to find potential errors. According to Mueller, current machine learning algorithms are mainly used to understand search requests and to generate new ideas.

Even if machine learning today plays an important role for the search engine rankings, there are many other factors that still have to be taken into account: especially the backlink structure and the relevance of a page.

Backlinks and the PageRank algorithm are the foundation of Google’s success. The Google index has evolved on top of the existing link network. Even if Google wanted to change this it would take a long time to do so.

One of the main principles of information retrieval is the concept of relevance. In other words the question is: “How well does the content of a document match the user’s information need?” This is and will be vital for the computation of the best search results. But Google and other search engines will become better in measuring relevance: The aforementioned interpretation of semantics by new algorithms like Hummingbird or RankBrain, user satisfaction metrics and feedback circles can help to separate more relevant websites from less relevant ones. Machine learning can support all that.

Conclusion

Machine learning already plays an important role in understanding search requests and finding relevant search results. With the evolution of server and cluster technology, the weight of machine learning will almost surely grow in the future.

After all, it is the users and their needs that will continue to take center stage of the search engines’ efforts. Hence user signals have to be taken into account for separating good results from inferior ones.