Artificial Intelligence Is Changing SEO -  Here’s How It Works

Over the past few years, the world has seen rapid advances in the field of artificial intelligence (AI). The advances haven’t just been limited to obscure, academic contexts but have had wide-ranging impacts that reach as far as content marketing.

Chief amongst these advances is the rise of what are known as transformer language models, the most popular being OpenAI’s Generative Pre-trained Transformer 3, or GPT-3. GPT-3 has heralded a new age of AI language generation — not just through the model’s power but also through the fact that it’s accessible for anyone to use. With its ability to generate high-quality text, far beyond any model that’s come before it, GPT-3 has had huge implications for the world of SEO and content creation.

What is GPT-3?

At its core, GPT-3 is an algorithm that tries to predict the most natural continuation of a string of words. If you feed it a simple prompt like “The capital of France is,” then it’ll respond with “Paris.”

‍

GPT-3 doesn’t do this because it has any innate knowledge of what France is, what Paris is, or what a capital is. Rather, it’s able to answer the question correctly because it’s been trained on the best part of a trillion words of text. As a result of this training, it’s able to notice patterns in language, which allow it to answer questions like the one above.

Historically, language models have been limited by the fact that they have to be trained on clean data, that is, data humans have manually inspected for spelling mistakes, formatting errors, and so on. In contrast, more recent advances in training methods meant that GPT-3 could be trained on huge quantities of uncleaned data and effectively teach itself to factor out these imperfections.

The latest GPT-3 models have been trained on data as recent as June 2021; however, the models are continually updated over time. This means that GPT-3 does face some limitations in its ability to talk about more recent events or concepts, which is something worth bearing in mind when using it.

In terms of the actual data that GPT-3 is trained on, 80% of it is what’s known as common crawl data, which basically means anything that you can find on Google. Much of this comes from well-known sites like Wikipedia, Reddit, or arXiv, but that also includes everything down to smaller blogs and informational sites. The other 20% of GPT-3’s training data comes from a variety of different sources but is largely made up of books and other long-form sources of content.

How is GPT-3 applied to SEO?

GPT-3’s real power, of course, isn’t in answering simple one-line questions but in producing responses to complex long-form prompts.

For example, if you prompt it with something like, “Write a very long description of the American Revolution,” you’ll get back several dense paragraphs of text about the topic. This is where we start to see how GPT-3 can be applied to producing SEO-ready content.

Some writers use GPT-3 as a writing assistant, to help in generating the odd bit of prose here and there. This use case reflects a lot of the tooling that has been built on top of GPT-3; in particular products like copy.ai and Jasper. These tools harness the power of GPT-3 to carry out certain tasks in the writing process, be it coming up with headlines, descriptions, blog outlines, etc.

While these tools have been impactful in speeding up content writing processes and reducing the costs of investing in SEO, they are somewhat limiting. For starters, it often makes more sense for writers to bypass these tools, and interact with GPT-3 itself through OpenAI’s playground (featured in the images above).

This not only saves on costs further, but it also gives writers complete control over how they’re interacting with GPT-3 and lets them tailor their inputs to their specific use case. This includes the ability to tailor parameters like tone of voice, length, and whether to focus on specific ideas in any content produced.

How to use AI to scale SEO

As previously mentioned, GPT-3 is capable of performing various different tasks in the writing process, including the writing of  headlines, article structures, and body content.

Where the power of this tool becomes  particularly meaningful, however, is when it comes to chaining the outputs of those different tasks together.

If you’re able to get GPT-3 to generate the article headline, a structure, and then content to insert into each section of that structure, then suddenly you have the ability to produce articles with minimal human intervention.

Each of these tasks is relatively simple to do manually; simply write out your instructions for GPT-3 with as much context as possible. For example, given the article headline above, we could get GPT-3 to generate an article structure as such:

And then for each subheading in the article, we could get GPT-3 to generate more content in the following manner:

Short and sweet, but you can always modify your prompt to produce longer paragraphs.

It should start to become clear that, by using GPT-3 to link up these different writing tasks, we can start to build a process by which it’s capable of automatically producing long-form, structured content.

It’s perfectly fine and well to create all this content  manually, rewriting your prompts by hand each time to produce each part of the article. If you’re technical, though, then it pays dividends to get familiar with GPT-3’s API. This will allow you to automate a significant proportion of the work to produce huge amounts of content in a fraction of the time that it would take a human to write.

How is this automation actually used by content marketers?

Increasingly, I’ve seen brands using approaches like the one above to automate a lot of their long-tail SEO work. While GPT-3 can’t compete with the depth and originality of a human when it comes to writing on important head terms (e.g., grow strawberries), more and more brands are using models like GPT-3 to build out rich content on long-tail search terms (e.g., how to grow strawberries in central florida).

Here’s why using GPT-3 for long-tail content works:

• Long-tail search queries often don’t need highly original answers, but they do need very specific answers, a use case to which GPT-3 is well suited.
• There’s a huge amount of variety in long-tail search terms, and most brands don’t have the budgets to pay content writers to cover them all.
• Search volume is often so low on some of those long-tail terms that it’s simply not economically feasible for brands to pay for human-written content. Whereas a non-specialist human might cost anywhere from $20 to $200, the costs of running GPT-3 to produce the same article are a mere fraction of that.

How can I use GPT-3 to build content for my site?

The sections above should provide some inspiration for you to get started using GPT-3, but there are a few additional points to consider.

It’s all well and good using GPT-3 to generate the odd article here and there, but if you want to use it at scale, then it’s worth a few considerations.

Choose to build content that:

1. GPT-3 is well suited to producing. This includes content that is matter-of-fact, is reasonably well covered (and therefore something that GPT-3 will have been trained on), and can be written without requiring accompanying imagery or illustrations.
2. Isn’t hugely competitive from an SEO perspective. The more competitive a term is, the more likely it’ll have been well covered by high-quality human writers. This doesn’t mean that GPT-3 can’t compete at all here; I’ve seen plenty of GPT-3 articles outrank well-written human content. Naturally though, your chances of success are best when targeting a large array of low competition terms, as opposed to a few high competition terms.
3. Follows a similar structure or format. If you’re producing lots of articles with a similar format (how do I do {x}, or what’s the best {y}), then you’ll be able to write prompts that are very specific and detailed but ones that also apply well to all of your articles. It’s much more difficult to write a series of prompts that work well across a diverse range of article formats.

A few specific use cases that I’ve seen work well are:

• Building glossaries. Most glossaries available online are fairly thin in terms of content, making it easy to outrank them with the rich content that GPT-3 is capable of producing. All the pages follow a very similar format (what is {x}), so you can be very prescriptive in how you prompt GPT-3 (write a set of factual paragraphs explaining the meaning of the term ‘{x}’).
• Building how-to guides. The sheer variety of questions that searchers can have around even a very simple topic can make it difficult to cover all the available search volume just with human-written content. This also means that these sorts of search terms are often less competitive and ripe for building out at scale with a tool like GPT-3.

The above should give some indications of the sorts of ideas that work well, though your imagination is really the only limit. If you can think of a large series of similar terms that your audience is searching, and which aren’t already super competitive, then you’ve likely found a good application of GPT-3.

What does this technology mean for SEO going forward?

While this doesn’t mean a huge amount for the core head terms that make up many brands’ SEO strategies, the next few years are likely to see a revolution in the quality and depth of content that’s available on long-tail terms. As more and more brands employ AI-driven strategies like the ones above, we’re likely to see those same brands capitalizing on the opportunities presented by being able to rank for their target audience’s long-tail terms.

While Google has issued warnings to sites that are using AI-written content, there’s little indication that they’re in a position to fight back. Their recent HCU (Helpful Content Update), aimed at penalizing AI-reliant sites, appears to have only hit sites with blatantly low-quality articles. All the brands that I’ve seen and worked with on AI-driven SEO projects have seen no impact on their rankings.

With that in mind, it’s easy to see why so many brands are employing AI models like GPT-3 in their SEO strategy — and why many more brands are likely to join the party over the coming years.