How Will Artificial Intelligence Affect SEO? Digital Marketing's 'Limitless' Pill

How Will Artificial Intelligence Affect SEO? Digital Marketing’s ‘Limitless’ Pill

Written by: Nicola Choi Posted on: 21st March 2018 Share this post:

There’s a (relatively) new kid on the block. And that’s artificial intelligence.

We’re not talking The Terminator’s Skynet, or Ava from Ex Machina, or any of those doomsday-style films. Fortunately, Arnold Schwarznegger’s time has mostly been taken up by telling people to claim their PPI refund. So we’re probably safe from a Terminator takeover.

Jokes aside, use of artificial intelligence in SEO and digital marketing is gaining traction.

And it opens doors for creativity and ideas galore. Why? Because, by taking the time spent by employees on these repetitive tasks and automating them, not only does it speed up the processes. It also lets everyone conserve a little bit more brain juice.

And that’s very necessary in today’s digital marketing landscape. Companies are stepping up their SEO game. Contacting SEO consultancies. Drawing up effective content strategies. Search engines using artificial intelligence (machine learning).

The competition is hotting up, and it was only about time before another big player came into the foray. And that player isn’t Neymar, no. It’s artificial intelligence.

So, is it really as good as it sounds? We’ll dive right in into the numerous reasons why AI might well be the key to making marketers even smarter in digital marketing and SEO this year.

We’re using Artificial Intelligence in our everyday lives, even if we don’t know it.

You know how you purchase a few items off Amazon, and all of a sudden, their recommendations for you are right up your alley? That’s Amazon’s machine learning from you. It’s tracked your order and purchase history. And it’s gone out of its way to find suitable, potential purchases that might catch your eye based on what you’ve bought before.

Using that information you’ve given Amazon, Amazon feels confident they could push another purchase from you by recommending appropriately.

Google’s RankBrain

Google’s, as expected, already on top of it. Currently, Google uses an AI feature called RankBrain. RankBrain is designed to deliver better and more relevant responses for queries. Though RankBrain is not the only factor or programme part of Google’s search algorithm, it does contribute to how Google ranks the results on its page, by relevance and reliability.

How exactly RankBrain work is still a mystery. The artificial intelligence component of it hasn’t been made explicitly clear by Google. But RankBrain could take into account factors such as browser history or the geographical location of the searcher in a bid to tailor results to them.

Moz, on the other hand, has given a bit more detail on RankBrain.

“The machine learning aspect of RankBrain is what sets it apart from other updates. To “teach” the RankBrain algorithm to produce useful search results, Google first “feeds” it data from a variety of sources. The algorithm then takes it from there, calculating and teaching itself over time to match a variety of signals to a variety of results and to order search engine rankings based on these calculations.”

We might not know exactly how RankBrain works, but we know what it does.

To give us a better idea of how RankBrain works, it’s probably best to try and put yourself in Google’s shoes. For example, if we were given the search query of ‘World Cup city’, how would you utilise RankBrain to produce the most useful result?

Something RankBrain can do that differentiates it from the rest of the pile is that it can try to determine the true intent of the search query. So, let’s step back into Google’s brain. What questions will we be asking to determine the answer to this question?

  • Does the searcher want to know about the upcoming World Cup? Or do they want to know about one of the World Cups in the past?
  • Do they want to know about the women’s football World Cup, or the men’s football World Cup?
  • Are they looking for directions to the city or a football stadium?
  • Maybe they’re doing a bit of historical research and they’re looking for the very first city to host the World Cup.

Because the search query is so vague, you can come up with all these questions.

In this situation, RankBrain becomes a necessity. Only by using its machine learning algorithm can RankBrain mathematically calculate accurate results based on patterns the algorithm’s noticed in searcher behaviour on Google. So, maybe people who enter that search term are actually looking for directions.

But if someone searched for the London Olympics stadium and their GPS coordinates correlates to someone living in London, Google may instead return a search result with driving directions. These kinds of factors — user location, the relevance and freshness of the content — are key to ‘read’ the searcher’s intent and satisfy their query.

A brief review of 2017 in Artificial Intelligence terms

In 2017, AI was a subject that was still a little up-in-the-air. Just out of reach of our comfort zones.

Why?

Not because it was good, but because it was really beginning. That’s in the digital marketing world, anyway. In 2017, AI showed us just how un-artificial it could seem sometimes.

Key examples include Amazon’s Alexa or Apple’s Siri. They both seem to have humour and personality ingrained into them in order to make them attractive. Furthermore, it almost makes them seem ‘more’ than ‘just’ an AI.

If we think about Siri, this is a computer programme designed to answer the questions we pose it in the most relevant, accurate way. You can even have conversations with Siri. How much of that screams ‘computer’ and how much of that screams ‘more than just a computer’?

Is AI going to be the patron Saint of SEO in 2018?

Maybe. Maybe not.

Some companies took the plunge and experimented with AI, reaping the rewards. Hopefully, with the success stories of the past year, 2018 will only become saturated with them. Companies now are aware of how to analyse their competitors using AI, and how to use that information to their advantage.

2018 opens up further opportunities to experiment in relation to digital marketing. One of the biggest hurdles you’d encounter is the public perception of AI. Understandably, some people may not want to wrap their heads around the notion of a computer automating everything.

Ultimately, there’s no denying that when programmed correctly, initially, AI can be an incredible resource. It may take companies and customers some time to develop a higher trust level with those two letters, AI. But it’s only early on in the year.

The rest of 2018 awaits.

Some suggestions on how to implement Artificial Intelligence in 2018

If the implementation of AI is a surefire thing, then 2018 is the year to challenge yourself, the company, to be more creative. Don’t be afraid to stick your neck out there with an innovative, up-to-date digital marketing strategy.

Some things don’t change. And this hasn’t. It’s about standing out from the crowd, having something different to offer, piquing interest where you might not have expected it from, etc. The purpose of your company via your marketing strategy may not have changed much, but the efficiency at which you can attain your goals certainly will with an AI under employment!

We’ve seen them here and there already. Some sites have chatbots that are operated under AI. For example, some customers might enjoy chatting online with an AI-operated customer service representative.

Even though simulated customer representatives might not be able to give your customer the exact answer, they can certainly funnel through options you can navigate through, and direct you to the most appropriate staff member.

Images

Images are already hogging the limelight.

We’ve got people sharing intimate details of their lives via pictures on Twitter and Facebook and Snapchat. Is there any part of someone’s life that isn’t documented now? And no, we’re not talking about the controversial surveillance state of George Orwell’s 1984 that we’re all now living under…

Whether it’s via the above platforms or as memes on social media websites like Reddit, Imgur or Tumblr, the Internet shares billions of photos and images every day.

In 2018, prepare for that data to be mined–by AI. It’s a well-known fact that a functioning AI can process information at such a speed that it’d put light to shame. But by mining all this data (in the form of images), AI can do so much more.

As technology develops, we doubt AI will just sit there with a library of cute puppy pictures, festering. AI is now used to help create customised content campaigns for each customer. Based on your browsing history an AI might be able to glean a better idea of the kind of photos you like to search for. In the future, it’ll deliver results via a faster response.

And have you noticed how when you search for something on Google, it comes up with related searches? That’s AI, too. The likelihood is, is that someone’s searched for those related terms after their initial query. AI can gather these, and suggest relevant terms for you that’ll help you find a more specific answer to your query.

Social media outreach

According to the Statistics Portal, 81% of the United States population has a social media profile. If we took this at face-value, that means, if you had a tailored, specialised social media strategy, you could potentially be reaching out to a whopping 81% of the United States population. And that figure will only be climbing.

However, there’s a downside to reaching out to such a potentially large audience. The huge amounts of unstructured social data, the number of interactions–let’s say on Twitter alone–could get remarkably overwhelming if you run a successful campaign

According to Lux Narayan, the CEO of Unmetric (an AI-powered social media metrics company), who’s been following this topic closely:

“I think there are distinctive patterns in what companies do. And indeed, in where they’re starting to deploy AI within each of those subdomains.”

So, we’ve outlined possible, perhaps even essential, applications for AI in the upcoming year.

#1: AI-generated content to partner your SEO strategy

This is an interesting topic to throw into the mix, if only because it can be potentially divisive.

If we consider regular, data-oriented events to be converted into news stories, that might fall comfortably within the AI writer’s territory. After all, there’s no bias to be reported. Just stats and figures. Blunt facts.

For example, there is an AI writing programme called ‘WordSmith’ that produced a whopping 1.5 billion pieces of content in 2016. It’s actually expected to grow further in popularity over the next few years simply because of the efficiency of their articles, and the quality (i.e. the way their articles sound as if they have been written by a human).

Surely they can’t write to a Dickensian standard?!

Of course, there is a downside. You can’t expect AI ‘writers’ to weigh in on public opinion posts, or a blog post advising the reader on the top tips from that industry. Those kinds of posts require, of course, human research, human empathy and a human opinion.

Could you imagine an AI writing down a political party lobbying piece? AI can’t give you that. It can churn out data at an eye-watering pace, but that’s why we need to be careful with what kind of articles you give your AI ‘writers’.

But when it comes to articles like quarterly earnings reports, sports match summaries, and other data, AI writers may be the way to go. Therefore, if you operate within a certain niche where the majority of your content consists of these kinds of articles, employing the use of AI may be well worth it.

Wait…If Content is King, then how impactful will artificial intelligence be in regards to SEO?

As discussed above, it’s entirely dependent on the type of content your site produces. If you are an opinion column, it makes sense to employ actual writers to get that across. Conversely, if you’re a company that utilises reports and they take forever to complete, AI will happily jump on that.

Collection of data, number, statistics and user or employee history is not time-staking for an AI. But it is, understandably, for a human.

So, on both hemispheres of the content world, there are advantages and disadvantages.

And finally, though it seems like an obvious point, it’s really up to you and how you envision your brand. If your company can utilise AI writers effectively, because your content is data-based and factual, it sounds like a perfect fit! But if your company churns out politically-minded columns and heavily opinionated pieces, with a personable voice, then maybe AI isn’t the best fit for your brand.

#2: Voice search

It’s estimated that by 2020, 50% of searches will be voice-activated. Going into 2018, there’s no sign of this slowing down.

Voice search is something that can heavily impact future SEO strategies. It’s something that brands should ideally keep an eye on and keep up with. If your company can respond to voice searches, you could gain potential gains in organic traffic, and that’s all thanks to AI-driven voice search traffic.

As a result, they’re changing how digital marketers seek to optimise their sites, and advertise. Sites aim to pull the customer away from simply browsing to an actual conversion. Though being top of a Google results page is important, it’s not the be all end all.

How many times have you spoken into your phone, asking where the nearest supermarket is?

Now, with AI assistants, they’re supplying the consumer with one, ‘best’ answer to their voice search query. For example, someone could say into their mobile phone: “Which Tesco’s is nearest to me?” or “What’s the top-rated restaurant in Edinburgh?” With the help of AI, the consumer will get delivered a satisfying, singular answer.

AI assistants are able to provide quick answers via complex algorithms. It can understand the customer’s query. Being able to distinguish between “who”, “what”, “where”, “how” and “why” has proven essential to voice search. The idea that people are relying on capabilities like this even more is becoming fact, as seen by the figure we’ve stated above.

More and more people are wanting a straightforward answer to their question rather than having to scroll through a results page. In a way, it’s almost like asking Einstein what the equation for special relativity is. You’ll get the answer straight away. He wouldn’t even have to think. That’s how quick AI can be, and that’s why it’s so often appealing.

Structured data: what is it?

How can you get on top of it, though, whilst keeping user experience at the forefront of your minds?

One way is doing this through something called structured data.

Structured Data

Data with a high level of organisation which search engines can effectively organise and display in creative ways. This helps brands in their quest to rank number one for specific keywords and attributes.

There’s very little room for error here. If the AI assistants provide a wrong answer, consumers are more likely to blame the business, not the device or search engine. Because information is easier to grasp nowadays, customers’ expectations have skyrocketed. Therefore, if you don’t provide the correct information quickly, it can result in a negative brand perception.

What does the AI do with this data?

AI relies on this data. So, it must be detailed, accurate and real-time. As more people rely on AI assistants, make sure you update frequently. Try to optimise your business locally–how many people these days ask their phone “what is the nearest X to me”?

If AI is relying on data, then maybe it’s time for us to rely on data too–structured data. Highlight key business information and create pages that are local-friendly, so customers can find them.

Rather, more importantly, the search engine can find them. Otherwise, how will your customer find them? Also, it’s useful to have a FAQ or query page where you can answer questions posed by customers. Not only will this boost your business’ image. But the more conversational tone you adopt in these FAQs for your target audience will resonate more with the person using their AI assistant to search for the answer.

#3: Predictive analytics

You might have seen predictive analytics at work on Amazon (a popular user of AI, it seems). Often, when you make a purchase, the next time you visit the website, you’ll be asked if you want to buy that particular item again. Furthermore, they often have items that are similar to the item you previously purchased, in hope that it’ll be suitable for you.

You might be wondering where the name comes from. It’s called predictive analytics because it uses the analytical data collected from you to make predictions about how customers behave.

Which customers will make repeat purchases. Which customers are likely to browse similar-spec items.

By doing this, it can make an almost ‘educated guess’ as to how to tailor its sales page to you.

If you’re a cynic like me, you may already be asking how?

If we turn this into an allegory, you can envision applying predictive analytics to a battle. Let’s say your AI has all the relevant data, such as the number of enemy troops, the terrain, the weather forecast, ammunition, etc.–then it can use that data to make this ‘educated guess’. Or, as we’ll call it, a simplified assumption.

Why is it simplified? Because as with anything, especially consumer behaviour–if we account for everyone’s individuality or ‘spurs of the moment’–you cannot cover all bases. In a battle, you cannot accurately cover or predict every factor that will impact the outcome of the fight.

For a consumer, you might have a collection of past searches and purchase histories, but you don’t know about changes in their life, and therefore the AI won’t know. They might have just redone their garden and on a whim decided to purchase a garden gnome. Predictive analytics can’t cover those random surprises.

More examples of predictive analytics include predicting the factors that’ll spur the customer into converting (e.g. price, the page of the website it’s on) or which customers may make repeat purchases.

Sounds perfect – are there any hidden issues?

As is the common problem with AI, predictive analytics is only as good as the data you collect. So, whilst there’s certainly an aura of Skynet fear lingering in the air, don’t worry. Applications like this–as clever as eBay and Amazon might seem–still need that human factor of double-checking and bias.

If there are errors made in the first place, or the data is sparse and random, then the predictive analytics won’t work very well, if at all. There’ll be no accurate predictions being churned out, and AI doesn’t have that capability yet of spotting that there’s something wrong in the source information and know how to correct it.

#4: Propensity modelling

Truly, this falls under the predictive analytics category. Essentially, a propensity model is what goes on behind-the-scenes. It’s how predictive analytics is applied.

Propensity modelling uses historical data gathered in order to form a model that can predict ‘actions’ in the real world. Using the propensity model in response to a query, predictive analytics can determine what the action to the query will be.

Propensity modelling is basically the end-goal of a machine learning project. The machine learning algorithm is fed large amounts of unstructured (historical) data.

Applying the model in the real-world

Theoretically, it can make accurate predictions about the real world. Overall, the goal is to study the features of item-buying customers, and then see how their behaviours differ from customers who don’t make a purchase.

Anything they’ve learned from this can contribute towards more targeted marketing, as well as creating, over time, a buyer persona for a particular item. This is particularly helpful for sites like Amazon. Or really, any shopping website. If you’ve ever wondered how Amazon always seems to know what to recommend in response to your search query, propensity models are likely behind this. For a diagram on how the model works, check out our picture below:

the propensity model of artificial intelligence

#5: Relevant lead scoring

Poor lead generation can have a massively negative impact on companies. On a personal level, personnel might miss sales target, or get demoralised. It doesn’t matter how good of a salesperson you may be. A customer who isn’t willing to buy won’t buy.

Lead generation is based on customer profiling. This is why your business will likely have client personas. That’s all the information you’ve gathered or know about your customers’ demographics, specific industries they work in, their job titles, their weak and strong points, etc.

Lead Scoring is a method used to rank prospects against a scale. This is a scale that represents the perceived value each lead contributes to the site.

The benefits of having a lead scoring system

Having a good lead scoring system is the best way to identify promising leads. By having an AI compare a list of potential customers to a list of current customers, the AI can automatically create a list of prospective clients by filtering out the ones who don’t fit the demographic.

According to a study on AI and consumer management: “66% of the 1,028 respondents were implementing or considering implementing predictive scoring technologies as part of their sales process. Of the 292 AI adopters surveyed by IDC, 83% reported that they used or plan to use sales and marketing predictive lead scoring.”

Tradition vs. the machine learning algorithm

Traditionally B2B marketers perform lead scoring manually based on a set of rules. If they meet certain criteria (e.g. two emails a week in response), they may be labelled a ‘hot lead’. So, you can manually differentiate between insignificant leads, and ‘hot’ leads.

Throwing an AI into the mix can help sites increase traffic and hopefully conversions by being quicker, broader yet more detailed. For example, it’ll use algorithms and predictive lead scoring to collect data on a user, and then use that data to predict the best lead score for this particular case. It can learn what made the user (or lead) shut down activity on your site.

When you understand and analyse this information retrospectively, you should surely be able to form a pattern of what parts of your site keeps your clients. And also, what parts of your site pushes the clients away. Furthermore, AI can help prioritise these prospects.

By increasing digital marketing Return on Investment and also allocating specific resources to aspects of the company/site that are known to work, AI massively boosts efficiency.

Efficiency boosted, what’s next?

Moreover, utilising AI will eliminate guesswork, and reduce the time it takes to gather and then analyse large amounts of data. So, even though you’ll still need someone to input the data correctly and then interpret it in terms of business, you can already see how AI will save you huge chunks of time.

But here, we’re moving forwards from predictive analytics to something called prescriptive analytics. Where predictive analytics gives a solid prediction on an outcome, prescriptive analytics provides insights on the recommended action that you could take as a result of whatever you’ve inputted.

Additionally, it doesn’t just provide information on what’ll happen. It’s scarily like a mini brain. It can also provide information on why something might happen.

How on earth can it see customer intent?

How? That’s the key question. Prescriptive models are built by not just analysing structured data, like the predictive model, but also unstructured data. This includes things like images, documents, and videos.

Because it’s essentially a black hole for all these types of information, organisations can use prescriptive models for a very scientific, pinpointed approach to their digital marketing strategy. This’ll result in better outreach and higher user conversion rates.

However, for companies to gain the most value from AI in terms of lead scoring, there does need to be adequate training provided for the digital marketing team. As we’ve said, the sterling results AI can produce can only be as good as the data you’ve inputted.

Of course, human judgement is always needed to add some context into why a lead converted. But with prescriptive models and the utility of AI, we’re edging closer towards that. Ultimately, it’ll result in better quality leads, saving a lot of time, and giving your staff more time to focus on other aspects of work.

#6: Targeted advertisements

Natural Language Processing (NLP) is a type of AI that’s proving to be popular in the digital marketing world. It finds pattern in online traffic and user behaviour. This means advertisers, because of NLP, can match their adverts with specific people based on several pre-empted factors.

One of the most famous examples of this is the Piccadilly Circus billboard in London that’s already using technology to choose what to display on its massive advert screen.

In this case, brands pre-programme their advert to show to target audiences. The algorithm that powers this advertising board takes into account gender, hair length, and height–amongst other factors–of passers by, and also cars.

For example, if it detected that there were a higher proportion of men walking by the sign, it might show a commercial for men’s aftershave. And this is just the beginning of slotting intelligent, targeted advertising technology in public.

But it isn’t just about reeling in present customers or passers by. It’s about the evolution of your brand and how you can cater your growth to your target audience. With the help of AI and its targeted advertisements, the hope is that the conversion rate will increase because you’ve deduced what the customer wants to see, rather than just another advert they’ll scroll past.

Facebook’s DeepText

Another problem marketers have encountered in the past is the ability to see things from the customer’s perspective, not just their own. This is a well-recognised hurdle in digital marketing. AI cannot read human minds (I don’t think so, anyway…) but with the programmer’s intent behind it, it can strive to…almost do that.

Facebook have introduced something called DeepText. As if Facebook didn’t know enough about you, DeepText can, among other applications, detect nuances or the ‘tone’ of language used in a post. It’s often been said that simply typing out messages wipes out the inflections and emotion in whatever you’re saying, but with DeepText, that could potentially be a thing of the past.

It’s good news for marketers. If the usage of DeepText becomes widespread, it might be possible to really increase the accuracy of an advert and who to target for it.

#7: Chatbots

Chatbots can be used to imitate human intelligence and can be used in many different situations. Mental health website 7 Cups of Tea use a chatbot in order to narrow down what the user wants from the service.

By asking a series of questions, the chatbot can determine why the user has come to the website, what kind of advice the user wants, and what troubles the user’s mind. All of this information collected by the chatbot is relevant due to the narrow, niche nature of the questions posed. Then the chatbot will hand the user over to an appropriate ‘listener’ who is trained to help with the specific problems outlined by the user.

In a business scenario, chatbots can even be used to take and complete orders from a website for a user or help direct that user to a specific place in an online store.

Facebook’s M

Again, Facebook can be pulled into the picture. With the development of Facebook Messenger’s own chatbot, ‘M’, Facebook wants to turn Messenger into the hub for people to converse with different brands’ virtual ambassadors.

(If you ask me, Facebook’s ‘M’ lacks the Dame Judi Dench’s magic. Just throwing it out there.)

Furthermore, they have developed technology, so you can use their service to train bots with sample conversations. The results will be open to you because you’ll be able to see how their bot’s responding to your queries.

Then, if you want to develop your own bot for your company, Facebook has written instructions for how to do so. The notion of bot creation being exclusive to big brands with big money is near fiction nowadays. On Facebook’s Developers site, they’ve created a tutorial for you to build your first Messenger bot.

Eventually, this’ll save you and your company a lot of time. As the popular saying goes, “time is money”. Efficiency can surely only be maximised if your qualified staff spend their time dealing with specific queries that need a human focus.

What are the ins-and-outs of a chatbot?

The automated ‘screening’ type questions, such as “how old are you?” or “are you a male or a female?” can easily be done via a chatbot. Gathering the answers to these questions, depending on the type of questions you set, will give the chatbot a rough idea of the customer. Therefore, it can accurately divert your customer to the appropriate member of staff.

For example, if the chatbot queried a customer and it turned out the customer wanted some help with building a website, it’d direct you to the company’s website developer, rather than the website’s content manager.

That way, it saves the customer time. They aren’t being passed over from one member of staff from another.

And also, it’ll save you time. Instead of spending time querying the customer for their needs, your chatbot does it for you. That way, your specially trained staff can focus on what they’re specially trained to do. Rather than handle general customer enquiries.

#8: Dynamic, automated pricing of products

As we’ve discussed above, using both predictive and prescriptive models to shape our AI friends can give them a slightly humanised voice. With everything being transferred online to a point where you can have a GP consultation online, marketing automation has become a $1.65 billion industry.

To say that AI could be a driving force behind that wouldn’t be too far of a reach.

If you’re selling a service, AI can be very effective. At a base level, us digital marketers know that sales equals shifting product. Yes, discounts can bring in more customers. But what profit margin are you left with at the end? And discounts are temporary.

Ideally, you don’t want a business that fluctuates wildly between successful and dead. For example, if you had a sale, the larger the discount, the more popular your sale is. But because the discount is so large, in some cases, you might even end up making less money.

Dynamic pricing of products can avoid this problem, and of course, this is where the AI comes in. The AI can create or suggest special offers only on those products that need them in order for a conversion.

How can this be applied in real life business?

Let’s put this in the context of a clothes shop. If you were to hold a half-price sale across your shop, but one of your purple jumpers sells an impressive amount anyway, you’d be losing money. Why? Because people are already buying that purple jumper.

On the other hand, if the sales figures for a sequinned dress are dismal, then dynamic pricing can target that dress. And it will decide that it’s suitable to be put on sale, because it’s not performing well outside of a sale.

Therefore, the AI can gather the unstructured data and create a propensity (predictive) model of whether the customer needs a sale to convert. Of course, for the purple jumper, we’d assume it doesn’t need an offer. For the sequinned dress that’s simply not shifting, it’s different. Maybe then we’d need an offer to try and push sales.

It seems easy enough for us to decide this. But if you had a clothes shop and you had to assess this information manually, it would take you forever. I’m not entirely sure if ‘forever’ is even an exaggeration.

Enter the AI.

And it’s not just general customer behaviour it can monitor, analyse and predict. It could also predict individual buyers’ behaviours. For example, it can monitor your data trail and predict the price you’re willing to pay for certain products. It’s seen in everyday life now. From Uber price surges to sites selling hotel or airline fares.

The Planet of the Machines

There’s always been a fear from humanity that one day, AI will pack its bags up, do a Terminator and blow us out of existence.

There’s always a ‘maybe’ in there somewhere when it comes to thinking about AI-related doomsdays. And it’s not helped by the warnings issued by world-famous geniuses like Elon Musk, or the late Stephen Hawking.

On principle, an AI is as good as what’s inputted. Yes, it has room for growth. Yes, it can evolve. Frankly, it’d be arrogant to assume that an AI can never match a human’s overall intellect. There are of course nuances to how we communicate and grow that AI finds itself lagging behind in.

But it definitely is safe to say that there’s a lot of bad press surrounding AI. There’s always the age-old question of “If you asked an AI to rid the world of poverty, would it choose the most efficient way and kill all people in poverty, thereby ridding the world of poverty?”

So yes, there’s ethics involved. And over the years, we’d don’t be surprised if the debate grows hotter and hotter.

But often, people simply just aren’t aware they’re already using AI technology already. As a digital marketing agency, understanding human behaviour has long been an obsession.

Why? I don’t think marketing our business’ use of Big Brother as particularly appealing…

By assessing human behaviour, we’re effectively assessing client behaviour. The ability of AI to transform such huge amounts of complex, unstructured, structured data–into something insightful, or to analyse it, means the possibilities are endless.

We can now technically assess the entirety of an individual’s online activity, from the emojis they send to the amount of time they spend on a certain page of a website.

When you throw in geographical locations, data from credit card transactions, store loyalty cards, spending patterns, browsing history…

Think of it as an exchange. Customers nowadays can access information in milliseconds. Why is that? That’s the result of years of human research. Now, AI has come along to help increase the accuracy of that research.

Are you saying that we really do live in an Orwellian state?

No.

There’s been hype (mainly negative) surrounding the idea. But using artificial intelligence in SEO can be hugely beneficial. And remember, for business purposes, it’s as good as you give.

As far-fetched and mind-boggling as AI might sound, the goal is ironically to draw us even closer to the customer themselves. It sounds like mass data collection. But that data can contribute towards us understanding, at a rather intimate level, the motivations of each customer. By doing that, we can give the customer exactly what they want. How they want it. And when they want it.

Perhaps the best thing to do is simply buckle up and prepare for the ride. Maybe the usage of AI isn’t as widespread yet, but 94% of marketers in this study said: “A tool that provides continuous, autonomous optimization across channels would be appealing to them”.

In the rapid modernisation of online marketing, all that’s left to answer, really, is what choice will you make? Red pill, or blue pill?

1 thought on “How Will Artificial Intelligence Affect SEO? Digital Marketing’s ‘Limitless’ Pill”

  1. This is a crazy topic just imagine if AI was able to rank your website within the day just because the AI learnt the algorithm of search engines.

Leave a Reply

Your email address will not be published. Required fields are marked *

Privacy Preference Center

Google

We use the following Google services to improve user experience and understand how we can improve our website.

Google Analytics,
Google DoubleClick

_ga, _gat, _gid, _gtld, _otpe, _ots, _otui

Close your account?

Your account will be closed and all data will be permanently deleted and cannot be recovered. Are you sure?