Get your head around AI

Bastian says: “It’s vital to embrace the broad minefield that is AI and machine learning. Fundamentally, this is an area that Google is investing lots into. It’s important to understand the landscape because you can’t optimise for something that you don’t understand. From an SEO standpoint, it’s a crucial driving topic for 2023 and beyond. That’s why it’s so important to get to grips with things now rather than being behind the curve.”

Are you talking about it from the perspective of Google using AI to better determine what content to rank?

“Yes. It’s an extremely wide area and AI in itself has different use cases. There are a bunch of tools out there where you can start inputting information around an article topic, for example, and receive insight. Whether you get great insight back or not is another question. This is a more hands-on use case, but when you zoom out you’ll see that most SEOs tend to like patterns and ranking factors.

AI in itself is a wider area, so Google is probably using it for research left, right, and centre. Its goal is to see where machine learning AI or AI support could make a difference in terms of better search results. It could then determine the content quality on a page and potentially be used to create content. It could also be used to decipher which content has been created by AI and what shouldn’t be ranking altogether. The future of AI is looking bright but, very soon, we’re likely to see parts of these implementations driving some of the rankings.”

Is Google using AI to determine what content is AI?

“Interestingly, yes. That’s the nature of the beast. You have this kind of ‘black box’ situation where people often struggle to understand what that really is and what’s happening behind the scenes. A lot of the current stuff we’re seeing is part of a machine learning algorithm, or multiple algorithms that work in unison to spit something out. We’re at a stage where the result of that is very much determined and driven by how good the training data is.

We talk a lot about English language stuff, but if you look beyond that we’re not quite there yet. The English language is one thing but beyond that, and even in other European countries, there is much smaller training data available. The output of that is way less in terms of quality, which is something to keep in mind.

If you’re talking about whether Google is there or not, this will probably only apply to the English search results. This is one of the reasons why, when you look at some of the recent announcements/updates, they tend to roll out in the English language markets first. Sometimes they’re rolled out in English exclusively because the corpus of data is much bigger than what’s available in other languages.”

Do you mean every country that uses the English language worldwide or perhaps just the US?

“Historically, Google has always rolled out in the US first. Right now, there are certain localisation aspects between the UK and the US. You could say the English language in general, even though there are different variations of it. Google masters English quite well but if it gets to more complicated stuff, or there’s less training data available, they might not roll things out in those different markets. That also means that machine learning can only do so much.

We need to move beyond that and develop a system that has some sort of intelligence to it. Once you have a system that isn’t relying on a limited corpus of training data it will be able to train itself. That’s why most people confuse the ‘machine-supervised learning’ versus ‘What is Artificial Intelligence?’ conversation. It’s often an oversimplification to call something AI when really it’s just a multitude of trained algorithms.”

Would you say that Google is actively using machine learning at the moment but they’re not using AI yet?

“Yes. That’s one thing that has changed historically, especially on the search quality side. They were previously very much against using machine learning in any way, shape, or form. That was a couple of years ago, but today’s visual platforms have many machine learning algorithms in place. For example, the recommendations we receive on YouTube. These are essentially all based on machine learning.

With AI, it’s a thin line, because it’s hard to judge. There’s been a lot of discussion around AI becoming this machine that has certain feelings. Google is using machine learning but not AI as much. There is an acceleration curve of the progress we’re making. Let’s look at how the role of the SEO has changed over the past five years vs what will happen over the next five years. With higher computing power and more data and capabilities to process that data, we’ll probably start seeing much faster progress.”

What precisely is Google looking at within the content to decide upon the quality and map intent to that? What do SEOs need to do about that?

“Historically there’s been a huge struggle between eCommerce businesses that want to sell products. They have to sell their products and have product descriptions around the items they’re trying to sell. However, often the decision has been formed much earlier, when someone is in an informational stage.

Let’s say you’re looking for a washing machine. Perhaps you have a bigger family now and want to know what type of washing machine you need. This is part of the informational journey. The question is, what do you really want when you search for a term like ‘washing machine’? Is that a purchase query or is it the information stage or somewhere in between? Conventionally you’d go to a local store seeking comparisons or recommendations. You could then attain search query refinement. Unfortunately, you won’t have the opportunity to have a dialogue with someone when you enter a search query.

That’s why Google is moving into this ‘user journey’ situation, rather than having one single input/output without understanding the context or meaning around it. From that perspective, what they’re looking for is to find results that can answer or support the user no matter what stage they’re in. It could be a piece of content that you get from a specific shop when the person is still in the informational stage of the journey. People might seek this comparison because they want to figure out which is the right product for them. You might then prefer to get a table or listing of different types of machines. You might want a single recommendation being shown.

Fundamentally, what Google is trying its hardest to understand is which companies or domains can help in all the different stages. This might not be true for every vertical but it’s especially relevant in eCommerce - where you have different types of queries supporting each other in a really important way.

Regarding holistic content, no matter how generic or specific the query is, Google’s biggest goal is to understand where someone is in the customer journey, and then serve the respective content. This could be very different when you search ‘washing machine’ vs when I search ‘washing machine’. The big thing is that the more data you have, the more machine learning you can apply - and the more factors you can get in. This is a crucial thing for Google to understand.

Let’s use searching for a new Audi car as an example. If you look at search results, historically there were times when they got a whole ton of brand rankings for their home and career opportunities pages. If you look at these generic search results today they’ve changed so much, because Google has got better at understanding what a proper selection looks like. For example, the searcher has an interest in driving an RS3 so we’ll serve this model page, not one that speaks about working at Audi because that’s not what the searcher wants.

If you turn it around and look at what SEOs do, it’s all about creating content that’s informing these different types of decisions and supporting different types of journeys. This is a big reason why there’s not one single answer to how long content should be. In previous years this has been determined by metrics - which aren’t reflective. The direction of content will be dependent on the topic and the amount of information that you can provide as a site. You should look at what kind of information you really need because, if it’s simplified information that can be served through structured data or with a single input-output answer, Google is going to do that themselves anyway. If you want to know how tall a certain building is, you can get that answer straight away. The biggest stuff that Google can’t answer straight away is where you need to build content that supports the entire journey. You need to support the entire process and serve the correct intent as an authoritative source.”

How do you become the trusted resource on a particular topic?

“Focus on EAT to become a trustworthy general entity. The entity could be a person or a multitude of people. This is a broad topic, but you need to build up a reputation over time because it isn’t something you can achieve overnight.”

Has machine learning radically impacted the way that authority is built?

“No, I wouldn’t necessarily say so. What it certainly does is help us understand if someone is, in fact, an authority because you can start mapping things out. If you’re in the pharmaceutical context, for example, it would be easy for Google to take a certain set of training data that’s solely related to prescription information for certain types of drugs. You could process that and - based on the specific corpus for a vertical like pharmaceutical - you’d know what a pharmaceutical site looks like, structure-wise.

It would be much easier for a machine to understand if it’s a typical site or whether someone is just writing about certain drugs for rankings. With machine learning, you can have verticalized approaches to using different types of data and therefore come to different conclusions as to what a specific authority in the pharmaceutical company looks like. Appeasing machine learning certainly helps but, generally speaking, becoming an authority doesn’t happen overnight. You need to invest in content reputation, citations, linking, etc. These things are still there but the process takes time.”

What shouldn’t SEOs be doing in 2023? What’s seductive in terms of time, but ultimately counterproductive?

“There’s still a lot of old-fashioned tactics out there. For example, auto-generated linking and these types of things. These outdated approaches have been dead for years, so let’s put them to rest. The general mindset you see is a desire to chase different types of patterns or isolated ranking factors that might only move the needle by 1%. This is something that people need to stop obsessing about because that’s not how search works anymore. If you have machines informing decisions based on a broad variety of data, yes you might move the needle by 1%, but the question is: is that really worth your time?

Don’t obsess over individual isolated ranking factors as much as you’ve probably done in the past. Yes, you’ll still need proper page titles but not only because they are a ranking factor. You need them because they make a difference in terms of CTR. They inform what that destination page is all about and could use an internal link, anchor text, etc. Don’t obsess over individual things and ensure your overall approach is as comprehensive as possible. Build a site that has meaning and is informing decisions along the user journey. Stop obsessing over isolated ranking factors.”

Bastian Grimm is CEO and Co-Founder at Peak Ace and you can find him at pa.ag.