While Google’s far from becoming Skynet just yet, as of 2015, it has officially infused its search algorithm with artificial intelligence (AI) – well, machine learning, at the very least.
First revealed via Bloomberg Tech, Google’s RankBrain, an AI machine learning system intended to help provide more reliable search results, has been in operation for about a year now. They tested it out for a few months, it surpassed their expectations, and now it is the third most important ranking factor in search engine results pages (SERPs).
Wait, Did You Say AI?
Don’t worry, it’s not the self-aware threat to humanity that makes Hollywood money and geniuses nervous. We’re still a ways off from True AI – an entity as intelligent (probably more) than man. What we’re talking about here is the application of algorithms that enable machines to make connections and “learn,” in a rudimentary sense. Machine learning and connected concepts such as deep learning are all AI systems that allow algorithms to not only process data, but use processed data in the future, in essence learning on the job, and then applying the learning to its next tasks.
Right now, the most widely covered machine learning system is most probably AlphaGo, Google DeepMind’s world champion-crushing Go-playing wonder. Before AlphaGo, everyone thought the board game Go was too complex for any AI system to learn, much less beat world champions at. Now Go world champion Lee Se-dol has been outmatched four games to one, an impressive victory that highlights the potential of machine learning AI.
So is Google infusing its search algorithm Hummingbird with the outstanding capabilities of AlphaGo? Not exactly.
RankBrain is meant to allow Google to learn about search queries that are initially unfamiliar to it. In their coverage, Bloomberg cited a search query example: “What’s the title of the consumer at the highest level of a food chain?” In this phrase, Google sans RankBrain presumably wouldn’t understand that the term consumer is meant not as a buyer, but as a consumer in the food chain. With RankBrain, Google not only understands the proper context for the word, but permanently learns the context and can successfully use it in the future.
All of the other ranking signals (more than 200 of them) are based on information and discovery, but not learning.
StoneTemple Consulting carried out an early study of RankBrain at work, picking out from a pool of 1.4 million queries those that indicate Google did not properly understand the query and that there were enough results for Google to find – the resulting set was 163 queries.
Out of the 163 queries that met their criteria, 89 were improved by RankBrain – basically, the machine learning AI improved results 54.6% of the time. In the example shown below, the query “why are PDFs so weak” was improved by RankBrain understanding the context of the adjective as it relates to the subject. In this case, weak security protocols for PDFs.
So, What Now?
It’s certainly an exciting time for technological development on the digital front, but let’s not lose sight of the practical ramifications.
First, as RankBrain continues to learn, so will it continue to become more useful. The 54.6% improvement rate shown by Stonetemple? That will only increase over time as RankBrain learns more and more. Great news for search users: more contextual, smart, rich results improving their experience similar to what Knowledge Graph has done. You can piggyback on this trend by doubling down on on-site optimization (e.g. get updated on HTML5 contextual codes), cranking up the rich snippet SCHEMA, and generally ensuring your target audiences’ experience from search to landing to conversion is awesome from start to finish.
Second, since RankBrain is the third most important ranking factor in Google’s Hummingbird search algorithm, which contains over 200 factors that themselves can have subsets of thousands of sub-factors, it’s inevitably going to affect your ranking. The most significant influence this will have is that as RankBrain learns, black hat tactics will increasingly become obsolete. There’s much more urgency now to abandon short-term algorithm-outsmarting strategies and start investing in long-term best practices.
Can’t really “outsmart” all this PLUS an AI that learns, like RankBrain (Image Source).
Third, only links and content are the ranking factors that come before RankBrain. Links will of course remain the hallmark of Google’s original PageRank factor, while content drives the linking behavior across the web. Since RankBrain is essentially an algorithm you can’t trick twice (not suggesting you try), the only way to increase rankings now is by targeting high quality links through high quality (and quantity) content. RankBrain has become Hummingbird’s smart, learning spam filter. No more shady shortcuts, just clean best practices.
Fourth: AI will inherently become a major part of online search. RankBrain isn’t the first. Microsoft has been using RankNet, its own machine learning system on search, in Bing since 2005. What this means is you should expect this to be the norm and use it in your strategy.
For instance, when Knowledge Graph became part of universal search, Google could show rich results based on the search query; see below for the query “Larry Page.”
One strategy to take advantage of this is to create so much brand recall online that Google will show your brand name and relevant links when someone searches for your brand or product line. For thought leadership positioning, you can create a lot of high quality, relevant content authored by your CEO to catapult him or her to “Knowledge Graph basis.”
RankBrain is pretty new, but some tactics are already being theorized. SEO Hacker assumed that since RankBrain learns offline – Google engineers “feed” it data instead of it actively learning online in real time – if you manage to coin terms that catch on like “growth hacking” or “inbound marketing,” this could become a signal of your authority in relevant subject matter and search queries.
AI systems like machine learning will have to be part of SEO as they have to be part of search, otherwise the stream of Big Data will overwhelm non-learning algorithms that can’t scale enough to crunch the information.
It really is all about holistic experiences now. When you’ve got learning search algorithms you know you’ve left the realm of keyword placement and density consideration far behind, and you should be ready to adapt your digital marketing to the new norm.
So, what significant changes should you make to your SEO campaigns given the rise of RankBrain?
Wordstream published an article focusing on click-through rates (CTR), analogizing that RankBrain behaves like AdWords’ Quality Score. Basically, it advocates to increase organic search CTR above industry average. Expect the industry to slowly start focusing on metrics other than the usual suspects in SEO, such as CTR, as Wordstream predicts.
Other than that, basically take all the implications discussed above to heart. First thing is start dropping short-term, quick gain efforts and instead use the time on long-term, complex strategies.
It’s no longer about creating a blog post about the single keyword phrase you picked from your keyword research. Now it’s about creating a blog post (or a series of them) around the concept related to that keyword. Obviously the tactic becomes more effortful and complex, because you’re not just leveraging the inner workings of an algorithm, you’re trying to satisfy a learning AI.
Furthermore, SCHEMA and rich snippet info will become increasingly important as RankBrain continues to gauge which results would be more useful to Google users.
Mind your linking behavior very carefully – clean up your link portfolio and start thinking less about link building and more about relationship building: it’s the relationship and context between linked websites that RankBrain values, not the single link exchanged.
Basically, generate more in-depth content and authority pieces, optimize the search experience starting from the SERPs (rich snippets), and watch your links. Finally, it wouldn’t hurt to update your previous content in important webpages to reflect these.
About the Author: Tim Clarke is the Research Manager with Clutch, Clutch identifies leading software and professional services firms that deliver results for their clients. Tim heads the SEO and PPC research at Clutch. You can follow Clutch at @clutch_co.