Natural language’s superiority to keyword-stuffed garbage when it comes to search engine optimization isn’t particularly new (though a lot of people in SEO seem to have become aware of it only recently).
Google has been mixing neural matching technology in their algorithm since at least 2018—and they’ve been using AI since 2015.
Using AI, neural networks, and deep learning technology lets Google’s bots understand just how relevant pages are to search queries through more than simple keywords, keyphrases (and yes, even key-synonyms).
Google’s AI now looks at a much bigger picture. Things like context, relation to broader real-world concepts—in short, naturalness—these all matter even more than they used to.
With the support of AI and neural network technology, Google is less manipulable than other search engines that don’t use their level of technology.
And since Google’s market share in search in 2022 is still right around 92%—a figure it’s stayed steady at since mid-2018—using natural language is no longer “one way of doing things.”
It’s now absolutely essential for any type of content marketing strategy.
Ironically, using artificial intelligence in its algorithm also gives Google an edge in that it’s far more likely to understand vague or malformed search queries that an actual human would not understand.
So in a sense, this AI is more intelligent than the average human—at least at understanding what someone might be searching for when they search using a mix of words and fragments of sentences, as most people today do.
Google’s bots aren’t the bots of yore that could be tricked by unnaturally repeating a bunch of keywords, bolding a few keywords here and there, and so on.
The Google bots of today have been upgraded with several different proprietary systems that allow them to interpret the way humans write—and the way we don’t write. Here’s a quick rundown on the advanced tech their bots are packing nowadays:
This system helps Google’s bots rank content through “deep learning,” interpreting how keywords actually relate to real concepts (rather than just looking for them and judging what percent of the content is made up of those keywords.)
Similar to RankBrain, this system looks at search queries and pages to identify the “human” concepts that are attached to them. In Google’s words, it’s different from their earlier AI in that RankBrain “helps Google better relate pages to concepts,” while neural matching “helps Google better relate words to searches.”
Is this confusing? Definitely. But if you just create interesting, useful content using natural language, you really don’t need to worry about the specific distinctions between these systems too much. Just know that they’re two highly advanced AI systems that Google’s bots use to detect whether your content is written naturally or if you’re trying to game the system.
Also known as Bidirectional Encoder Representations from Transformers (we’ll just use BERT going forward), this is what Google uses to understand how the meanings of words change when they’re used in different ways and different orders.
This is why typing “WHO” immediately pulls up the World Health Organization, typing “a Who” pulls up Horton Hears a Who, and typing “The Who” pulls up… The Who.
Google’s Multitask Unified Model is essentially a more ominously named, much more powerful version of BERT. Like 1000x more powerful.
With MUM, Google doesn’t “merely” understand human language—well, actually 75 human languages—but images as well.
All four of these systems currently work together to help judge your content and rank it accordingly when someone does a search.
Keyword density is dead. Google’s bots are much more intelligent than they were 5 years ago.
They know how to tell whether you’re a human writing real words—or not.
And they’re not going to get less intelligent in the future.
You can bet that, as you’re reading this, they’re hard at work creating new natural language detection systems that are even more capable than MUM right now.
After all, if these are the ones they’re telling search engine marketers all about, what kind of technologies are they keeping hidden for now?
There really isn’t much you can do to “game” these systems going forward—beyond creating real, high-quality, naturally-written content. And that’s a good thing.