“Open the pod bay doors, HAL.”
“I’m sorry, Dave, I’m afraid I can’t do that.”
This iconic movie exchange from 2001: A Space Odyssey was science fiction at the time. Today it’s not.
It can be easy to forget, but every time you ask your phone for directions, a machine is understanding what you say. That’s a very new phenomenon.
Google has gotten pretty good at figuring out what you want, and there’s a reason for that. The secret is BERT, or “Bidirectional Encoder Representations from Transformers.” BERT is changing the way Google works — and you need to understand how so you can create content that delivers.
The Rise of Natural Language Search
Natural language processing, or NLP, is the process of building technology that understands and responds to human input like text or voice. BERT is a natural language processing (NLP) model.
So in layman’s terms, Google’s building a search engine that understands your intentions and doesn’t just match words or phrases to a database.
This is a bigger deal than it sounds. For most of search engine history, you had to have a good grasp of methods like Boolean search syntax to get good results. Yahoo, one of the foundational early search engines, functioned like a card catalog, with manual categorization and descriptions for every site in its directory. WebCrawler, released in 1994, used the first automated crawler to index pages, but it still wasn’t that “smart.”
That changed when AltaVista launched in 1995 and was the first search engine to allow natural language queries. Google’s predecessor BackRub launched in 1996 and rapidly overtook AltaVista, growing to the giant we see today. Its relevance-based algorithm changed the way we use search.
The next major leap forward came in 2010, when Google’s voice search and Siri both hit the market. Voice search meant that Google was suddenly dealing with a huge volume of natural-language queries, and a wider variety of query wordings. They were heavily invested in NLP at that point. And with the release of BERT in 2018, Google sprinted past all previous NLP iterations.
You might be surprised to realize that one in ten Google searches currently uses NLP. It adds context to searches, using both the string before each word and the string after each word to determine the meaning.
Take Google’s example, “I accessed the bank account.” Bank is the word we’re trying to find the meaning for. Many NLP models would use “I accessed the” to determine the meaning and usage of bank (institution, not river). BERT uses both “I accessed the” and “account.” Sounds simple (and there’s way more to it than that), but it makes a world of difference for your optimization.
What Does BERT Mean for Your Content Marketing?
I’m personally fascinated by AI, machine learning and the future of the web, but you’re not here to hear me go down that rabbit hole. All you need to know is what does BERT mean for you as a content creator?
Google search is learning what really matters to people. This isn’t a new thing; it’s been broadly apparent since the Hummingbird update in 2013. Every technical advancement Google search has taken since it was founded is aimed at one goal: how can we make Google search give people exactly what they want?
The algorithm has traditionally worked like a vending machine: punch in a specific input, get a specific output. Google is trying to make it a conversation.
Search right now is a zero-sum game. If your content ranks and you get a clickthrough, you win and everyone else on the search engine results page (SERP) loses.
With the model Google is working towards, there is no “top of the SERP.” There is “top of YOUR SERP.” Each user’s search journey is going to be different, because Google can use search history, NLP and other tech it’s working on to parse meaning — just like you would in a conversation.
The Content Conversation
Copycat content (content that looks similar to everything else on the SERP) has been a useful SEO tool for a long time. But as Ryan Law over at Animalz noted recently, Google is actively trying to get rid of it. You need to be adding something new to the conversation. Just publishing something because “it has the right keywords” and “it has good backlinks” isn’t going to work anymore.
When the top pages that come up for most searches are long, high-quality “skyscraper” pieces aggregating every possible question you could ask about a subject, that’s an improvement over the bad old days of search. Google incentivized that type of content. But once you’ve read one of those pieces, you’re much less likely to engage in follow-up searches or click on anything else on the SERP. There’s no diversity of information. That’s why “information gain” is such an important concept for SEO moving forward.
NLP and other enhancements to Google’s algorithm are creating a different SEO future. When Google is getting a huge variety of natural-language queries and giving personalized answers, the prize goes to the page that answers the question that was asked the best – not the one that wrote a generic piece about the keyword.
Just like you would in a conversation, you have to find something new and interesting to add. You can’t just repeat other people’s ideas in your own words, or the conversation will die (in Google’s world, you’ll get penalized).
Want to Google-proof your content? Answer the specific questions your audience is going to ask; don’t just think about keywords in isolation. Google isn’t going to be a machine you can “game” for much longer when it’s getting 5.6 billion new data points to learn from each day.
Science Fiction is Now Science Fact
The world where your computer understands what you have to say isn’t science fiction anymore. It’s here. And it’s advancing every day.
And as content marketers, that’s great for us. Because the closer Google gets to understanding what you want, the more great content wins. Take the time to create content that goes deeper than “me too.” Say something interesting and new. Answer your audience’s questions and BERT and other NLP models will serve you up to them on a silver platter.
Rainmaker Digital Services