AI SEO Trends: To Love AI or Fear AI, That is the Question

Let’s face it, AI is unavoidable. 

Whether it will completely “kill SEO” or be a process we all adopt is still to be seen, but there’s no denying that it’s almost impossible to avoid. 

Source: Portent

AI has caused plenty of controversy in SEO over the last few years, with constant fights over using AI vs not using AI. 

A few months ago, Jake Ward (a well-known SEO on LinkedIn) posted an AI SEO case study on how his AI tool, Byword, was able to pull off an “SEO heist” using competitors’ sitemaps.

His case study became an instant hot topic in the SEO industry with him being covered by a few well-known publications and even receiving a few death threats for it (talk about negative digital PR).

Most will agree the case study was in poor taste, but I would hope everyone would agree that violent threats are a bit too far. 

But for the future of SEO, should we fear what’s to come with the all-knowing AI, or should we embrace it for what it is? 

With this article, I’m hoping to explain where I think AI is headed and if it’s worth worrying about just yet. 

(Maybe Not So) Brief AI Recap: Where We Currently Are

*You can skip this section if you want to get straight to AI and SEO. This section will just talk about the history of AI. How it started to where we are now*

While AI might seem like “the new thing,” it’s actually been around since the 1950s. Alan Turing published the first document “computing machinery and intelligence” which opened the doors for AI.

For the purpose of this article, I’ll be talking about large language models (LLM) and generative AI.

But what actually is AI in the context of generative AI and LLM’s? Google describes it as:

Artificial intelligence is a field of science concerned with building computers and machines that can reason, learn, and act in such a way that would normally require human intelligence or that involves data whose scale exceeds what humans can analyze. AI is a large field that includes many disciplines including computer science, data and analytics, software engineering, and even philosophy. 

At the business level, AI is a set of technologies that has many use cases, including data analytics, predictions and forecasting, natural language processing, recommendations, machine automation, intelligent data retrieval, and more.

While we still haven’t reached a fully self-aware AI (although that’s not completely out of the realm of possibilities according to OpenAI), most AI we talk about now are LLMs that have been trained on a large set of data.

This quite literally comes from a large set of data, as IBM puts it:

LLMs are a class of foundation models, which are trained on enormous amounts of data to provide the foundational capabilities needed to drive multiple use cases and applications, as well as resolve a multitude of tasks.

An LLM is a deep learning algorithm that can perform a variety of natural language processing (NLP) tasks.

If you aren’t famixliar with natural language processing, NLP combines computational linguistics, machine learning, and deep learning models to process human language.

And if you don’t know what deep learning is, this is how Google explains it: 

Deep learning is a subset of machine learning that uses artificial neural networks to process and analyze information. Neural networks are composed of computational nodes that are layered within deep learning algorithms. Each layer contains an input layer, an output layer, and a hidden layer. The neural network is fed training data which helps the algorithm learn and improve accuracy. When a neural network is composed of three or more layers it is said to be “deep,” thus deep learning.

Deep learning algorithms are inspired by the workings of the human brain and are used for analysis of data with a logical structure. Deep learning is used in many of the tasks we think of as AI today, including image and speech recognition, object detection, and natural language processing. Deep learning can make non-linear, complex correlations within datasets though requires more training data and computational resources than machine learning.

And as mentioned, deep learning is a subset of machine learning. So what exactly is machine learning?

Here’s Google again to help us out:

Machine learning is a subset of artificial intelligence that enables a system to autonomously learn and improve without being explicitly programmed. Machine learning algorithms work by recognizing patterns and data and making predictions when new data is inputted into the system.

LLMs and generative AI might be commonly associated with tools like ChatGPT, but the roots of language models and natural language processing have been around for quite awhile, again going back to the 1954.

In the Georgetown–IBM experiment, researchers developed a system to automatically translate a collection of phrases from Russian to English.Using a set of 250 words and six grammar rules, the machine translated sixty Russian sentences into English. 

The earliest prototypes of language models were rule-based systems. These systems, such as ELIZA (1966) and SHRDLU (1970), relied on a set of pre-defined rules and lexicons to generate and understand language. ELIZA, for instance, was an early Natural Language Processing (NLP) computer program that emulated a Rogerian psychotherapist by rephrasing many of the patient’s statements as questions and posing them to the patient. It used pattern matching and substitution methodology to simulate conversation, but it didn’t understand the conversation.

SHRDLU, on the other hand, was a system developed to understand and respond to commands in a restricted world of geometric shapes. It was one of the first programs to demonstrate natural language understanding, but its capabilities were limited to the specific domain it was designed for.

While these tools may have been revolutionary at the time, they still had significant limitations. 

Again in the 1980s, IBM started early development of the first (small) language model. This model they built was designed to predict the next word in a sentence. Part of their design includes a “dictionary,” which determines how often certain words occur within the text the model was trained on. After each word, the algorithm recalculates statistically what the following word should be.

During the 1980s and 1990s, statistical approaches became more prominent in NLP. Researchers relied on Hidden Markov Models (HMMs) and n-gram language models to predict the probability of a sequence of words. The n-gram model used co-occurrence frequencies of words in a dataset to make predictions.

Recurrent Neural Networks, or RNNs, were created in 1986 by researchers David Rumelhart, Geoffrey Hinton, and Ronald Williams. They developed a special kind of neural network that remembers its previous inputs by looping information back into itself. This makes RNNs really good at understanding sequences, like sentences in a conversation or steps in a video, because they can remember what happened before and use that information to make sense of what comes next. 

Long Short-Term Memory Networks, or LSTMs, were introduced in 1997 by Sepp Hochreiter and Jürgen Schmidhuber. They are a special kind of Recurrent Neural Network (RNN) designed to remember information for long periods. The main issue with traditional RNNs was that they struggled to keep track of information if there was a long time gap between important data points, because they tended to forget earlier inputs. 

LSTMs solve this problem with a unique structure that includes ‘gates’—components that control the flow of information. These gates can decide what information should be kept or discarded, allowing LSTMs to retain important past information and forget the irrelevant, making them much better at tasks like speech recognition or language translation.

The 2000s saw the rollout of neural networks. 

One of the first and most notable neural language models was Bengio et al.’s Neural Probabilistic Language Model (2003). According to Level Up Coding, this model introduced the concept of learning a distributed representation for words and a probability function for word sequences. It was a significant departure from previous models, and it set the stage for the development of more advanced neural language models.

During 2013, Google’s Word2Vec was a significant breakthrough in NLP, using neural networks to create word embeddings that captured semantic meanings based on contextual usage.

Word2Vec takes as its input a large corpus of text and produces a vector space, typically of several hundred dimensions, with each unique word in the corpus being assigned a corresponding vector in the space. Word vectors are positioned in the vector space such that words that share common contexts in the corpus are located close to one another.

Word2Vec was groundbreaking in capturing semantic and syntactic relationships between words, including analogies (e.g., “man is to king as woman is to queen”). 

In 2014, Google released the Seq2Seq model based on Long Short-Term Memory (LSTM), which helped for tasks like machine translation.

In 2014, the concept of the generative adversarial network (GAN) was presented. GANs are used to create images, videos, and audio that seem like authentic recordings of real situations.          

Then with the rollout of the Transformer model, as introduced in the groundbreaking paper “Attention is All You Need” by Ashish Vaswani et al. in 2017, which was considered to be a significant turning point in the field of NLP. This model introduced an architecture that was based entirely on attention mechanisms.

The attention mechanism, specifically the self-attention mechanism used in Transformers, is a method that allows the model to weigh and prioritize different parts of the input sequence when generating each word in the output sequence. Scaler describes this as allowing the model to “pay attention” to different parts of the input depending on what it’s currently processing.

This mechanism is particularly effective in capturing long-range dependencies in text. Traditional models like Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks often need help with this as the length of the sequences increases. 

However, the attention mechanism in Transformers allows the model to focus directly on any part of the input sequence, regardless of its distance from the current position. This makes it possible to handle larger sequences of text and capture dependencies that span across large gaps in the input.

The Transformer model has been the foundation for many subsequent models in NLP, including BERT, GPT-2, GPT-3, and most modern language models. These models have built upon and extended the Transformer architectures. 

OpenAI’s Generative Pre-Trained Transformer (GPT-2), released in 2019, was one of the more well-known LLM products. With 1.5 billion parameters, it was a substantial leap from its predecessor, GPT-1, which only had 120 Million parameters.

Building on the success of GPT-2, OpenAI released GPT-3 in 2020, which now has 175 billion parameters. 

DALL-E is a machine learning model created in 2021 by OpenAI that generates photorealistic images from text descriptions. 

Following the release of GPT-2 and GPT-3, OpenAI introduced their most well-known product, ChatGPT. This model was specifically designed to generate conversational responses.

This brings us to 2023 with GPT-4, with an insane 1.76 trillion parameters. While GPT 4 has been around for some time now, GPT-5 is soon to come.

And lastly in our little history lesson, we have Google’s Search Generative Experience (SGE), which is an experimental update to Google’s search engine that uses artificial intelligence to generate contextual answers to search queries. Almost like a ChatGPT built into Google’s algorithm.

SGE has officially been rolled out in 2024 and is now going under the name AI overviews.

Its reception by the public has been very mixed so far.

AI vs SEO: What’s the Outlook Here?

So, where are we headed? 

Should the SEO industry embrace AI, or should we be worried about what’s to come?

Despite what many on either side say, no one knows. 

No one can predict the future.

However, there are a few trends we can point to that might give us an idea of the future to come with AI. 

Using AI for Content Creation: Is It a Yes or No?

As mentioned in the intro, AI is a wildly hot topic among those in SEO. 

Mainly when it comes to content creation.

Some think it’s the best invention since sliced bread, and others think it’s the devil and should be avoided at all costs. 

I’m here to argue that both sides are somewhat wrong. 

At least in AI’s current state. 

It’s still way too early to predict where AI is going, but it’s only going to improve. 

That much is true. 

But in its current state, how should SEOs be treating AI?

Well, for starters, AI has its pros and cons. 

And before I dive in here, let’s clear the air about something real quick  – Google has explicitly stated that they are fine with content however it was created as long as it doesn’t violate their spam policies and is still relevant to the searcher. 

Now, an important thing to know here, Google doesn’t have the technology yet to truly understand what is good content and what isn’t good content. 

They have a series of patents in place that allow them to connect different words to understand the meaning of a particular piece of text. That’s also why entities have become so popular in the last few years. 

They can’t gauge the quality of the text, but they can gauge the information covered in the document using entities and connected words.

That’s why information gain has been such a popular topic recently. 

While Google is unable to interpret the actual quality of your content, they can gauge the information coverage found within your content. 

From Figure 5 in Patent US20200349181A1

So, if your information is just repeating all the other rankings articles, Google will likely be less interested in your content.  

But if you’re adding something new to the conversation, your content will have more perceivable value and relevance attached to it. 

Now, there’s a big difference between word count and relevancy. 

Just adding more content/information doesn’t automatically make your content rank better. 

Your additional content has to be relevant to the overall topic.

It’s like creating a 5,000-word chicken parm recipe vs covering additional deduction tips to a tax-related article that other articles haven’t covered. 

These tips will likely come from personal experience and provide additional value that other articles haven’t covered.

Think about it like this: why should Google prioritize your content if it’s just rehashing what’s already out there?

Site authority (not DA) plays a huge role in this also but that’s a whole other conversation. 

But why am I going on this long rant about information gain when this article is about AI?

It’s because of the common misconception that Google can interpret your content’s quality and therefore your AI content won’t rank because it’s naturally low quality. 

The answer lies in the information and relevancy of your content.  

Just because it’s AI doesn’t mean it won’t rank, but at the same time, randomly publishing AI content won’t guarantee rankings either. 

In fact, that one might actually get you into trouble with Google’s spam policies or their new helpful content system, which was designed to demote unhelpful content. 

As we’ve seen with the SEO heist – which has since been issued in a manual penalty during Google’s March Algorithm Update. 

In fact, the image below is of Casual, the site used during the infamous “SEO heist.” Despite being targeted back in December, the site still hasn’t recovered.

Google even updated their spam policies to target “scaled content abuse” which are sites scaling content without providing value to users. 

AI content is a common culprit of this with sites like Byword (Jake Ward’s tool) allowing you to publish 100s of articles with a click of a button.

This update only made the AI/content debate more contentious, with 100s of sites being issued a manual penalty/deindexed for abusing Google’s spam policies.

So again, why am I going through all this? 

It’s because AI content can create helpful content. 

You just need to know how to create helpful content with it. 

In its current state, the proper way to create content using AI is to take it one step at a time.

You basically need to prompt each section of your article vs having it generate it for you all at once. 

So, using AI to generate content can be worth it; you just have to know how to use it. 

And more importantly, be patient with it. 

It’s also worth adding that AI outputs aren’t always accurate and can even lead to hallucinations.

With AI in its current state, you should still fact-check and review your content so you’re not misleading readers. 

Simply put, have at least some human editing process with your final output. 

While Google may not be able to gauge “quality,” do it for your readers. They’re the ones reading your content after all.  

SGE’s Rollout: Is There An Expected Onslaught or Not?

Probably the most likely to “disrupt” (not kill) SEO is Google’s Search Generative Experience (SGE) feature. 

This has also been a hot topic due to the quality of the results being shown and in some cases, directly displaying content from other sites without properly citing their source. 

That’s been my personal gripe, along with others in the SEO industry. 

SGE can be a helpful product, but in its current state, it’s practically stealing content from site owners without giving them proper credit for it. 

So where is SGE likely headed now that Google is starting to officially roll it out?

My personal thoughts are that SGE won’t be as successful as many think. 

As mentioned before, SGE in its current state, isn’t adding anything new to the conversation and is generally just rehashing what’s already directly below it on the SERPs. 

This will probably change since it’s still technically a young product, but it’s very disappointing in its current state.

However, there is one thing to consider with SGE. 

SGE will probably wipe out the need for some queries. 

Primarily top-of-the-funnel informational queries. 

Searches like me looking for direct information like “when the final tax date is.”

I don’t need to read a 2,000-word article when I can get the direct answer in 2 seconds. 

But even with these queries being impacted, I’m still optimistic about the rollout of SGE and SEO.

If anything, it makes me more optimistic about SEO. 

Even though SGE will likely steal away those queries, those queries likely weren’t even valuable to us in the first place.

Informational searches primarily worked well for brand awareness but usually didn’t serve much of a purpose outside of that. 

Now, will SGE affect non-ToFu searches? 

Maybe.

But AI will never have personal experience. 

It’s built on large sets of data, as we discussed earlier. 

There are just some searches that require expertise to answer them. 

Only an expert or person knowledgeable in that field can offer their own insights to help a searcher.

This is truthfully the main reason why I’m bullish on SGE and SEO in general. 

It’s almost going to force us to create better content due to a new playing field. 

No longer can you just put words on a page and have it rank. 

To actually get our audience’s attention, we need to create content that is genuinely helpful. 

Subject matter experts are going to be huge in a post-AI world. 

There’s a reason why searches have been geared more towards Reddit recently; searchers want first-hand experience, and our job as SEOs is to make that content accessible for them. 

In my opinion, SGE and Google’s HCU are flushing out the previous tactics that gave SEO a bad name like keyword stuffing and rehashed content. 

Sadly, there were some indirect casualties, but hopefully Google corrects that and continues to head down this route.

The main complaint with Google and their SERPs is that they’re filled with useless information that all rehashes the same thing. 

If we want to stand out, we have to create content that genuinely separates us from the other results. 

Google describes it as “people first content” but this even ties in with my first point. 

Google is working towards showing content that’s genuinely unique and relevant to the searcher and we’re just lending them a hand with getting them there.

SEO has stayed the same and isn’t dying; it evolved. 

So with SGE, while yes it will put a hurt on some searches, I don’t expect it to hit queries that really mattered to our sites anyways. 

Hurting How Clients View SEO and SEO Partners

The biggest challenge I see SEOs facing is taking on client expectations due to AI. 

And it’s only going to get worse from here on out. 

AI is being pitched as the ultimate scalability tool, a cheap assistant that does everything for you at a fraction of the cost. 

While this is partially true, I can see clients willing to test this out for themselves to avoid paying for SEO services. 

If they know how to do it, then great. But as mentioned before, the large majority will not know what they’re doing and might even end up doing more harm than good.

This even goes back to the first point about using AI for SEO content. 

Yes you can do it, but mindlessly generating hundreds of articles in the name of SEO is only going to hurt your website. And actually, hurt your site in the case of a manual penalty. 

SEOs can maybe see this as a plus since it will likely result in plenty of recovery situations, but I think I vouch for most of us when we don’t want to see clients fall into that trap anyways. 

The main concern I have here is that clients will think AI gives them the ability to fire writers and SEOs. 

If AI can help with scalability, then great! But I think we all know it’s a risky move.

Rehashed Information and Misinformation Everywhere, You Can’t Avoid It

This ties mostly into point #1 but I think eventually AI content will turn into a negative spiral. 

Large Language Models are trained on large sets of data, and that data is usually taken from websites. 

If everyone is using AI to create content, then the AI will continue to generate the same generic output without adding anything new to the conversation.

It will eventually just rehash the same information over and over again. 

This is obviously an extreme example, but from what we’ve seen from AI so far and how easy it is to abuse, it’s not too far off from reality.

If we thought the SERPs were bad now, just wait until every article is just rehashing the same content over and over again. 

The main theme of this article is new information, and the same exact thing applies here too. 

Since AI can be easily abused in its current state, I think this is a destination we’re steamrolling towards awfully fast. 

Where We Might See Some Benefits from AI 

Okay, while I had plenty to discuss on the potential cons of AI, there are a few pluses worth highlighting. 

Not everything has to be negative all the time.

Making Scalability More Accessible For Site Owners 

It’s hard to argue the benefits of AI, especially when it can be a solid tool to help with accessibility and therefore, scalability.

As long as AI is being used properly, it can be a solid tool for scaling up your content creation process. 

It’s not a tool that should be used to replace, but a tool to be used to elevate.

Things like:

• Writers using it to speed up their ideation process
• Site owners using it to help with automation
• Writers creating content faster
• SEOs being able to use AI at scale (create MDs, product descriptions, etc.) 

Less Tedious Work

Piggybacking off the previous point, AI will help business owners find new ways to pull less levers. 

Tasks that used to be ignored can easily be automated thanks to AI. 

Example: creating meta descriptions or product descriptions at scale. You won’t need to take writers away from creating content when you can generate it quickly with a click of a button. 

As is the theme with this article, this only works if you’re editing the output, not just copying and pasting. 

But even with the human involvement, it’s still saving you plenty of time. 

A New Way To Pick Our Brains

To me, this is the biggest benefit AI has. 

It can cut down significantly on the time for research (not keyword research) to expand or improve your content. 

Just a note that AI does tend to hallucinate a lot or produce junk outputs so it’s up to you to weed out the useless noise or false information. 

But putting that aside, it’s a great way to pick your brain and expand on areas of your content you might’ve missed. 

Your writer can use it to:

• Research areas of their content they might be missing
• Find statistics that can make your content stand out
• Create rough drafts of content
• Put together quick research blurbs on content

Dr. Strangelove or: How I Learned to Stop Worrying and Love AI

There are so many different angles to AI that I could talk about this forever. 

But is all the worrying worth it? 

My answer after 4,000 words is….maybe. 

But most likely no. 

I wrote this long article talking about the pros and cons of AI and still landed here; why?

AI is still very much in its infancy phase so all we can do right now is speculate. 

We have no idea where AI will be in 5,10, or 15 years. 

It can all change very fast. 

ChatGPT was only released 2 years ago.

However, the reason I’m optimistic is due to the need for expertise and experience. 

AI might have access to more data, but it will never have the personal insights of a person deep in the trenches. 

A contract lawyer will have years of experience (data) that AI will almost never have access to.

Those insights will only become valuable as AI becomes more accessible. 

My personal take is that AI will only continue to be abused in a large-scale race to the bottom, which will only make it easier for others to differentiate themselves.

So, for that reason, I’m going to leave the big answer at a maybe.

But ultimately, I have a more optimistic view of where things are headed with AI. 

How Can We Learn to Live with AI

AI is here to stay, and there’s no denying that. 

What we should do is learn to live with it rather than choosing to ignore it. 

We need to recognize its benefits while also recognizing what its clear limitations are. 

Simply put, use it, but don’t treat it as the holy grail. 

How AI might be used with SEO in 5 or 10 years, no one has a definitive answer yet.