When Google says that it is an AI first company, it means that all of its products have AI at their core. Specifically, it means that all of its products have Google Assistant at their core!

Google’s Sundar Pichai, says that he doesn’t want Google to simply use AI in its hardware and software, but rather to invent products and services from the ground up that couldn’t exist without AI.

A perfect example of this is Google Clips, a camera that you wear that takes photos automatically when it thinks something exciting is happening.

This is the kind of concept that is only possible thanks to machine learning and computer vision.

Even the Pixel line of phones can be seen primarily as an attempt to put Google Assistant right into people’s pockets.

And this is where the real “big idea” makes itself known. Google wants to get to a point where all of us talk to it every single day, using natural language. Google Assistant has the ability to answer questions about nearly any given topic, by understanding what the user is asking, looking for relevant information on the web, and then reading that back in an easy to understand manner.

Because if AI really is going to be the next HUGE thing on the web like everyone is saying, then Google wants to make sure that it is the go-to platform that everyone uses. This is Google betting BIG on the future, and going all in on an AI focussed line of products and services.

So what does all this have to do with internet marketing?

The thing to recognize here, is that Google Assistant works using Google Search the two products are in fact so closely linked as to be almost one and the same. Google Assistant is ultimately a ver y fancy form of voice search, and Google Search is essentially an assistant that you talk to by typing.

For a while now, Google has been leaning toward a more natural language type of search.

Back in the old days, Google worked essentially just by looking for matches between search terms and keyphrases.

If someone searched for a specific phrase, then Google would look for content that used that precise term.

But this had problems. For one, it made it extremely easy for marketers to “game the system.” You c ould manipulate Google’s algorithms by inserting keyphrases repeatedly, even if your content wasn’t of any interest to the person looking for it.

At the same time, it meant that high quality content that didn’t happen to do keyword research wouldn’t get anywhere.

And it meant that Google would often serve up the wrong content. It would look for keyword matches, but with no regard to the sentence. If you searched for “decision tree” then it might bring up content about making decisions about trees rather than the flow chart meaning that you meant.

Google adapted by introducing RankBrain an algorithm designed to better understand what the user actually meant rather than what they said.

RankBrain can do a few things. For one, it takes words and segments them: categorizing them into phrases that it thinks might be related to one another. This helps Google to better understand what a user wants to learn about, and it means that Google can guess what a word it isn’t familiar with might mean.

Those groups of words (word vectors) are categorized using a process known as distributed representation. What that means, is that words that are close together in terms of meaning and context are grouped, thereby helping Google to understand more and allowing it to be more flexible in the way that someone searches.

RankBrain will then try to map the query into words that it can understand and will look for related terms and phrases.

These will then be used to sift through the huge indices of content that Google has access to, and to find the most relevant results for the user.

In the past, Google wanted you to search terms like:

“Buy hats online”

But now it wants you to speak naturally using a phrase like:

“Hey Google, where can I get nice hats online?”

Thanks to RankBrain, it will know that in this context, “where can I get” is essentially equivalent to “buy.” Thus the user feels they can speak to Google more like a human, and that they will get more relevant results than ever before.

RankBrain is also able to understand the relationships between words, and that includes the role of words like “and” and “or.” These can subtly change the meaning of a search term, but previous versions Google’s algorithm would simply have ignored those joining words.

RankBrain also improves on its own results over time. It can use data such as CTRs in order to ascertain whether it provided the correct results, and use this information to inform itself for next time.

It’s constantly improving.

RankBrain is so advanced, that it actually needs specially designed chips called Tensor Processing Units (TPUs) in order to work. When you search for something on Google, it will actually outsource the question to these TPUs in the cloud and use that to sift through huge amounts of cont ent.