Artificial Intelligence In Digital Marketing

Being smart in business means knowing what’s just around the corner. It means thinking ahead and preparing for inevitable changes that will impact the way business is conducted. This is what allows a business to be resilient and to thrive in a changing environment. Digital marketing is no different. In fact, in his book The Personal MBA, author Josh Kaufman discusses the value of counterfactual simulation. This means imagining future possibilities and then preparing for them. Let’s say that you have a big business that is doing well in a specific niche. 

Maybe you have a company that sells a whey protein shake. The mistake that some big businesses make is to assume that they’re too big to fail and to coast along as they are. But what would happen if another company came along and released a better protein shake for a fraction of the price? What if a new source of protein were to be discovered? What if a study revealed that whey protein was bad for us? Any of these things could happen, and could completely shake up even the most established business. The smart company though, will already have considered these eventualities and prepared for them. THIS is counterfactual simulation: it’s thinking about what’s just around the corner and then preparing for those possibilities. As digital marketers, that means thinking about things that could impact on the face of marketing. And one of the things that could have the biggest impact of all? Artificial intelligence. 

AI and machine learning have the potential to completely change the face of internet marketing, rendering many older strategies obsolete even. Only by preparing for those changes, can you ensure that your websites manage to hold their position in the SERPs, that your advertising campaigns remain profitable, and that your services remain relevant. And a lot of this stuff isn’t just speculation: it’s happening right now. AI is already making huge waves even though you might not realize it yet. It’s affecting the way that SEO works, the tools and software we use, and the way that ads are displayed. AI is able to think faster and smarter than any human, and that’s especially true when it comes to internet marketing which is a data driven pursuit. An AI marketer can create endless amounts of content in a second – doing the work of hundreds of humans. All of that content will be perfect catered toward the target demographic. 

AI will run Google. It will manage entire business models. It will run AdWords. And it will run new tools we haven’t even dreamed up yet. The digital marketing singularity is just around the corner. This book will help you to prepare, and explain a number of 

concepts:

• AI vs Machine Learning

• How to conduct SEO now that Google is an “AI first” company

• Chatbots

• Programmatic advertising

• Big data

• RankBrain

• Digital assistants

• Data science

• SQL

• Latent Semantic Indexing

• The future of internet marketing

In this post, you will gain a crystal ball with which to gaze into the future of internet marketing, and to 

ensure that you are ready for all those changes when they come. By the end, you’ll be better prepared 

and in a better position than 99.9% of other marketers.

Before we go further, we should first take a look at precisely what AI and machine learning actually 
are. These are two related but also distinct terms, which often get confused. Both will impact on marketing, but in different and unique ways.
AI then is artificial intelligence. That means software and hardware designed to act and appear intelligent. Such software is capable of making meaningful choices, and conducting activities that we would normally consider the remit of humans. AI comes in two broad flavors. One is weak AI, which is also known as narrow AI. Weak AI is essentially a form of AI that is designed to perform a specific job. An example of this is the self-driving car. This form of AI is capable of knowing the positions of countless cars on the road, and being able to respond by steering, accelerating, breaking etc. If you were to watch a self-driving car from the outside, you might think a human were driving. In that way, it does a job that would normally be considered a human role. BUT at the same time, you can’t speak with a self-driving car and you can’t ask it how it’s feeling. A self-driving car would certainly not pass the Turing Test!

Weak AI might not sound as exciting, but it is being used for a huge range of extremely exciting things – from helping to treat disease, to improving the economy. Conversely, the type of AI that we often see in science fiction, is what we know as “general AI.” This is AI that doesn’t have just one purpose, but that is designed to do everything that a human might be able to do. So you could play a word game with this AI, ask it how it’s feeling, or get it to look up something useful. An example of a general AI is DeepMind, owned by Google. DeepMind is a company that has developed a “neural network,” that employs “general learning algorithms” to learn a huge range of different skills. Many AIs such as IBM’s Watson are actually pre-programmed. That means that they work using a kind of flow chart, and will answer questions with the same answer every time. On the other hand, DeepMind is apparently able to think and respond via a “convolutional neural network.” Certain behaviors and reinforced and encouraged, and these will begin to become more prominent. This isn’t a perfect simulation of how a human brain works (cognitive behavioral psychology teaches us the importance of having internal dialogues and models for thinking), however it is the closest thing we currently have to a “true” general intelligence. 

Machine Learning 

Machine learning on the other hand works differently. Machine learning utilizes huge data sets in order to gain surprising and almost frightening capabilities at times. Machine learning essentially allows a piece of software to be “trained.” An obvious example of this would be computer vision. Computer vision describes the ability that some machines have to understand visual information. An example is Google Lens, which can tell you what you’re pointing your phone’s camera at, whether that’s a type of flower, or a product you can buy in stores. Computer vision is necessary for self-driving cars to successfully navigate their environments, and it’s used by apps like Snapchat which use filters to change people’s faces. How do these work? By looking at thousands and thousands of pictures of every type of object. 

While the machine learning algorithm will never understand what it is looking at, it can look for patterns in the data which will then be useful to identify those objects in future. For example, it might notice that faces are typically oval in shape, with a dark patch of hair on top. It then knows that if it sees an oval shape with a dark patch at the top, it’s possibly looking at a face. Machine learning has HUGE potential in just about every field. In future, it can be used to diagnose disease more accurately than a human doctor, to advise on financial decisions, to identify fraudulent bank transfers, and much more. All of this has HUGE potential implications for internet marketing, and that’s what we’ll be exploring in the following chapters. 

A while ago now, Google announced that it had become an AI-first company. While that might sound like meaningless marketing babble, the truth is that this determination actually has HUGE potential repercussions for marketers, businesses, and SEO. Firstly, what does Google mean by this? Meet the New, Smarter Google You might think of Google as a search-first company. The first product that Google provided was a search engine and this is still what most of us associate with the company. Traditionally, Google’s search engine did not work much like an AI. Rather, search worked by attempting to match search terms with the content in an article. 

This is why the advice for SEOs was to insert lots of key phrases into their articles, so that Google’s spiders could read that content and quickly identify that it would be a good match for what the person was searching for. As we all know, this didn’t work out perfectly for Google. Lots of unscrupulous “marketers” abused the system by inserted hundreds of search terms into every article, which in turn meant the content Google would show to the user would be garbled and unreadable. That’s why, over time, Google has begun to work more and more like an AI. Now, Google no longer attempts to look for exact keyword matches. Instead, Google tries to answer questions that you ask it. It does this by trying to understand what the user is looking for along with the context, and then to provide relevant answers through its search. 

RankBrain 

Google is able to do this through machine learning. Specifically, it uses a form of natural language processing, which Google refers to as RankBrain. RankBrain is at least partly responsible for helping Google to cope with phrases and words that it hasn’t seen before. If RankBrain identifies a word it isn’t familiar with, then it can “guess” what it might mean based on context and based on its usage elsewhere. This helps Google to deal with unusual searches that it hasn’t seen before, without simply matching search terms to content in articles. Search queries are turned into “word vectors”, called “distributed representation.” These are words and phrases that are close to each other in meaning and context. RankBrain will then try to map the query into words it understands, or clusters of similar words. From there, it insinuates what the searcher actually means and is looking for, and provides results on that basis. RankBrain also understands the relationships between words, and the way that they work together. At one point, joining words such as “the” or “and” were ignored by Google. 

Now Google understands the importance of these phrases and the way in which they impact on the intent of the user. Like all the best machine learning algorithms, RankBrain attempts to improve over time and adapt to users. It can see which results get clicked the most and thereby know when it is doing well and when it is getting things wrong. As such, it is able to improve search results for any given keyword quickly through algorithmic testing, which is helping to weed out low quality content that attempts to game the system. RankBrain works using a Tensor Processing Unit (TPU), which is an AI specific piece of hardware stored in Google’s data centers. This is a specific chip that is better able to handle the specific challenges of machine learning tasks. 

Google’s Further Plans 

Over the past few years, Google may have seemingly diversified. It now makes smartphones, it now makes self-driving cars, and it now makes apps like Google Lens. But at the heart of all of these initiatives is some form of AI or machine learning. Google Lens uses machine learning to identify objects in a scene and allow users to that way “search” the real world around them. Self-driving cars of course are highly reliant of various forms of AI. And the Google Pixel Phones? Arguably, their main focus is putting Google Assistant in everyone’s pockets. And this is the real clue as to what Google is up to. Google Assistant is an AI and virtual assistant that users can use to get weather reports, to book taxis, to play music, and much more. Google Assistant uses a combination of machine learning (to detect human language for example) and AI in order to provide useful results and speak in a natural manner. Google Assistant is closely integrated with Google search. 

You can ask Google Assistant a question like “who starred in Iron Man?” and it will give you a natural answer. It does this by first using machine learning to turn your speech into a string, then by using Google Search in order to look up useful answers (which involves machine learning in the form of RankBrain), then by using narrow AI to extract the most useful answers from the best web pages, and then by using another form of narrow AI to provide the response in a natural-sounding manner (which is designed to appear like general AI.) Much of this is carried out not on the device that you’re speaking to, but on Google’s TPUs located in the cloud. 

What Does All This Mean for Marketers? 

So what does all this mean for marketers? Simple: it means that Google wants to be able to understand your content and extract the most useful information. It no longer wants you to use rigid keywords, and it wants you to prepare for a more voice-driven form of search. Google is betting big on AI and machine learning. It believes that in the future, AI assistants will be HUGE and it wants Google Assistant to be number one. It envisages a future where we spend less time staring at our devices and instead get the information we need by asking our phones or our Google Homes. We’ll speak naturally to these devices, and they’ll provide us with handy answers. 

Whether Google Assistant eventually becomes the ubiquitous tool that Google wants it to be or not, the fact remains that Google wants search to become increasingly more natural and human. It already has in many ways. That means that marketers and website owners need to make some changes to the way they do things. It’s no longer enough to find a keyword and repeat it a whole lot, you now need to work as though you’re speaking with an AI. And that means a couple of things. 

LSI: Latent Semantic Indexing 

Latent semantic indexing is one of the most important things to consider if you’re interested in improving your SEO and getting to the top of Google. It’s even more critical if you hope to be ready for Google’s AI-driven future. Not only is it a powerful concept in itself, but it is also an important microcosm of the broader changes that we are seeing to SEO today. Search engine optimization is a big and very important part of digital marketing and if you want to drive the maximum number of people to your website or blog then it's absolutely essential that you have the search engines on board. In the past, SEO has largely relied on creating tons of content around a certain topic and repeatedly using a set number of keywords or key phrases in that content in order to help Google identify the subject and help the right visitors to find your pages. 

Unfortunately, a few people began to take advantage of this system and began 'keyword stuffing' by using the same keywords over and over again to the point of distraction. Google had to get smarter and so it did. Today, using the same keyword too much will get you into trouble. So what does Google do instead? It looks at context and the broader subject of the article. In other words, it looks for synonyms and related terms and this also gives it the ability to better understand what your page is about. For instance, if you had written an article about “decision trees,” then in the past Google could theoretically have gotten confused and brought your site up as a result when someone searched for trees. It may have thought you were talking about decisions about trees! Now though, it can look for related terms like “flow chart” and thus help to more accurately match article to reader. LSI actually comes from mathematics, and uses a technique called singular value decomposition. This means that it will scan unstructured data and look for the relationships between the words and concepts within.  

How to Handle LSI 

So how do you make sure your site is LSI optimized? Short answer: you don't. While it is obviously tempting for SEO companies to now start offering their LSI optimized services, the truth is that you should have been doing this all along and without thinking about it. That's what the best web marketers like Andrej Ilisin have always recommended and its what Matt Cutts advises as well. In short, writing naturally should mean that you are including synonyms and related topics. Otherwise your writing is going to sound pretty repetitive. The moral is what it's always been: stop double guessing and just write for the reader! This is something we’ll come back to again and again with regards to preparing for a smarter Google. But there are also some other tips you can keep in mind if you want to ensure that Google knows what you’re talking about. 

First, make sure that you use more than one search phrase. It’s a good idea for a whole host of reasons to use a combination of different search terms, rather than targeting just a single one. Seeing as Google will often show results that don’t use the exact key phrase the person searched for, it makes sense to try and include a few popular iterations of the same term. Likewise, you should make sure to use good and varied vocabulary around the topic. This helps to better demonstrate the context and the subject matter of your article. Rather than filling an article with random synonyms, think instead about words that would often occur alongside the topic you’re working with (such as our earlier example of flow charts.) This is called co-occurrence, and it’s the kind of thing that machine learning algorithms love! 

Structured Data 

The other big concept that SEOs need to consider in order to be ready for the AI Google of the future, is schema markups, also known as structured data, also known as rich data. Remember: Google’s aim is to enable Assistant to answer natural language question with useful responses, which will draw on information found on the web. Google doesn’t just want to pull up a list of useful search results, it wants to be able to answer questions. So if someone asks how to make bolognaise, it will simply read out the ingredients. In order to do this, Google needs to be able to find that most relevant information in a passage of text, and then pull out the specific answer. This takes the concept of RankBrain to the next level, allowing it to understand not just what an article is about, but how each paragraph in that article functions. 

The problem is that Google’s AI can’t quite do this yet. At least not well enough to be able to usefully provide answers for people without occasionally including completely nonsense! That’s where schema markups come in. The idea of a schema markup, is to essentially annotate your articles and blog posts by telling Google what each bit is and what it does. Essentially, you are saying “this is a list of ingredients” or “this is a user rating.” This also helps Google to provide what are known as “rich snippets.” Rich snippets are search results on the SERPs (Search Engine Results Pages) that include more than just a meta description. You might see a search result listed for instance that also includes bullet point steps, or that includes ingredients for the meal. This way, the user can see the information they’re looking for without even needing to leave that website! 

How to Use Markups 

Markups look a lot like HTML. Here’s an example of what this might look like:

THE CANDLE FACTORY

888-888-8888

That is basically telling Google that you are talking about a local business (The Candle Factory). You can also use schema to highlight product names, authors, aggregate ratings, software applications, restaurants, movies, and much more! To use this yourself, you can either look up the HTML code and implement it yourself, or you can use Google’s handy markup helper: https://www.google.com/webmasters/markup-helper/u/0/ 

Here, you will simply share the URL of the page you want to markup, and it will then provide you with the opportunity to create the necessary tags. There are also plugins you can use to the same end through WordPress. 

You might hear the term “big data” thrown around a lot and not fully understand what is meant by it. In this chapter, you will be enlightened and learn how big data can help you and your business, both now and in the future. Essentially, big data is nothing more than large data sets. These large data sets are increasingly common online, seeing as everything online is easy to measure and document. If you think about a company like Google, it has immense data sets that it works from, describing the search history of billions of users.  But even a standard website that gets 1,000 visitors a day will work with huge amounts of information. 

 A website will naturally record each of those visits and will also store data about each one – such as the country of origin, and the length of time spent on the site. In a few weeks, this data will likely crash a lot of spreadsheet software! The reason that big data is featured in so many discussions is that it is very difficult to handle. Making sense of such huge amounts of information requires a lot of smart math, while simply storing and handling that kind of data requires a lot of storage and computational power. But the potential value of big data is also absolutely huge. Big data provides patterns and insights that you simply can’t get by observing a few users. This is essentially how machine learning works – by looking for patterns in massive data sets. The difference is that this is being leveraged in a slightly different way. 

Predictive Modelling 

Predictive modelling is a process that involves data mining and probability to forecast potential future outcomes. A model is created using a number of “predictors.” Predictors are variables that are thought to influence future results. Once data is collected for those predictors, a statistical model can be created. That might use a simple linear equation, or it might use complex neural networks. Either way, statistical analysis can then be used in order to make predictions about how things are likely to go in future. With regards to marketing, this can help provide better customer insights, better lead scoring, campaign nurturing, upselling and cross-selling, personalized product recommendations and more! Amazon is an example of a site that uses big data in order to provide personalized product recommendations. 

Amazon doesn’t just use a database of items grouped together (which would be almost impossible to maintain) but rather generates data automatically from every single transaction and sale, and then looks for patterns. It will see what products tend to be bought together (there’s that co-occurrence again) and can therefore use this information to show items that it thinks a user might want to buy next! Likewise, when it comes to lead scoring, big data can be immensely useful. Lead scoring means understanding which leads are likely ready to purchase and which are not. This is immensely useful information for companies that might want to send sales letters to the cross section of their mailing list that they think will actually buy from them (rather than being put off by the amount of sales material they’re receiving). 

Amnesty International uses segmentation and “predictive modelling” techniques in order to better identify the right groups to market toward. By collecting data and then looking at what that data reveals about the kinds of people who donate, Amnesty International knows who it should be targeting with its ads, how much they are likely to spend, and how they’re likely to do it. Any charity can benefit from this kind of data analysis, as can any business. 

Collecting Big Data 

If you want to start collecting data for your business, there are a wide number of plugins and tools you can use to do so. You should find that a lot of tools, such as Google Analytics, will allow you to export massive amounts of data in order to work on. You can then choose to use this information yourself, or to outsource it to a data science organization that can use that information to provide valuable, useful insights. Another good idea to prepare yourself for the future, is to allow users to create profiles. By doing this, you can collect much more data on individual users, and in future provide better recommendations on an individual basis too. This is something that stores have been doing for decades with loyalty cards, but of course the digital nature of selling online creates even more potential opportunities!

As mentioned, computer vision is the ability for machines and computers to “see” by learning from huge data sets and machine learning. By observing countless images, a machine can learn to identify images in an object, or to navigate an environment without crashing into things. What does this mean for the future of SEO? One BIG thing – and one thing that you should make sure that you are ready for – is that Google will likely start paying more attention to images on websites. Traditionally, we have been told to avoid using images for things like site names. Why? Because Google can’t “read” and image, and therefore we won’t get any SEO benefit from that. 

 But Google does have software that can read text from an image. This is called OCR (Optical Character Recognition) and if you want to see just how good it is, try pointing Google Translate at a foreign language and see it appear in your native tongue in real time. If Google can do this, then it’s only a matter of time before it starts reading the text in your images to see if they back up the niche and keyphrase that you are targeting with your website. Likewise, seeing as facial recognition is already a big deal when it comes to security and Facebook, it is probably only a matter of time before Google starts using that too. For example, if you write a blog post about Sylvester Stallone, Google might someday look not only at your content, but also at the photos on your page in order to see if there are pictures of Stallone there! 

 Google Images might one day not be reliant on surrounding text at all, but might instead base its results purely on what it sees in the image, and whether this lines up with what you’re searching for. Issues like image quality are also likely to play a big role in future. Google might opt not to recommend your webpage if it thinks the imagery on there is poorly chosen and out of place. So what can you do to prepare? For now, the closest thing to communicating with Google via images is the use of markup language and/or file names and alt tags. Using alt tags to describe images can help Google to know what they represent, and therefore to better decide if your site is relevant for a particular user. Meanwhile, make sure that all the imagery you are using is relevant and high quality! 

>>>Part 2