Focus on your audience to avoid getting bogged down

Emina says: “First of all, don't get bogged down with all of the flashy terms.

We're all talking about fan-out, chunking, and cosine similarity. All of that is amazing, I love technology, but also go back to the basics of marketing and who your audience is.

Those tools are brilliant as a way to help you understand your audience better, but they are not the goal in themselves. Understand the basics first.”

How do you define who your audience is, and why is that important?

“It's important because it all comes from knowing your audience.

During my degree in development studies, I had a great professor who talked about sanitation aid programs that were going into developing countries to build toilets, and they were failing to get people to use them.

Then, somebody came up with the Community-Led Total Sanitation (CLTS) approach, which is a program where you go into the community and you do social mapping to understand where the rivers are, where they go to the toilet, and all of that kind of thing. You facilitate a discussion. That organisation then had a really high success rate; they were building toilets, and these communities were actually using them.

That’s a great example for marketing and SEO. As marketers, we tend to go from top down: ‘This is my audience. This is what I want them to do. They're going to have to deal with it,’ but that’s not the case. You need to build the toilets the way that they want the toilets built, rather than you telling them.”

What does the process of building a website with this approach look like?

“If you’re building a website the way that Community-Led Total Sanitation would do it, you would go and do user testing, user research, and market research.

First of all, you would start by understanding the company. Let’s say you're an in-house digital marketing manager, and you've been tasked with redeveloping the website. First, you have to understand what the business is, from the inside. Talk to customer service, sales, and anybody who can give you an insight into what they think the business does and how it does it.

Then, you go externally to your audience to understand who the audience is. 15 years ago, I sat at the University of Sussex doing usability testing based only on Steve Krug's book. I sat there with wireframes of my new website and tested it among students because they were my target audience, and I wanted to see whether or not I was on track.

You can do it face-to-face, but you can also do it really well with LLMs. This is where cosine similarity, fan-out, and all of that come in. This is where your audience is, and you have LLMs to help you crunch that data and understand it. You have APIs that allow you to bring that into your local environment really quickly. All of that is amazing, rich data about your audience.

Then, you go and build the toilets in the way that the audience wants them, rather than from the top down. You build your website the way that your audience wants it, for the way that they search, and in the way that they understand.

Those sanitation programs had been happening for years and years, but they just weren’t working. Then somebody realised the reason why it wasn't working was that it wasn't starting with the community. It's the same with marketing; you've got to start with the audience.”

Why do you believe that audience research is the foundation of SEO in a zero-click AI search reality, and how do you measure the impact of that work?

“This is the marriage that absolutely has to happen between SEO and digital channels, and conversion rate optimization. Yes, you have less traffic, but the traffic that you do have, you should aim to convert. In a situation where you have less traffic, the traffic that you get is even more precious because you want them to convert.

Ultimately, what we want is not traffic; it's conversions. It's revenue, not empty visits to the website. This is why I'm not super paranoid about traffic going down, because a lot of our clients are not seeing the impact on conversions. That is partly because we've been working with them on the CRO piece as well.

You need to rethink your KPIs. Your KPI isn't traffic, it’s revenue, it's conversions, and it's engagement metrics. That is really important. How you do that is by making sure that your website uses things like Hotjar and other tracking software. Make sure that you're looking at your user journeys and you have the analytics to support you in that.

Start with your measurement framework. In a situation where there is no traffic, what do you need to measure?”

Can you explain what the term cosine similarity means?

“Cosine similarity is the one that everybody knows. It's basically a way of understanding context. Let's say, for example, you have two documents: one of them is about jaguar, the big cat and one of them is about Jaguar, the car. If a user simply types in ‘jaguar’, how does a search engine or an LLM understand which one they're talking about?

One of the ways of doing that is through cosine similarity. I'm massively simplifying here, but it is basically calculating the cosine angle between all of the vectors within the corpus of that document. If it sees a sentence that says, ‘A jaguar, the spotty orange cat, was running through the woods,’ that’s obviously talking about the big cat.

The term ‘jaguar’ is there, but it also has ‘cat’. Sometimes it's not even that obvious. It might say, ‘A jaguar is spotty and it has fur.’ There's no ‘cat’ in there, but it will know that you’re probably not referring to a car because you have ‘fur’ and you have ‘spotty’. Cars are not furry.

Whereas, if you have a document that talks about ‘mileage’, ‘engines’, etc., then it's obviously about a car. It's a way of understanding the semantic relationships between words in an embedding space.”

What is query fan-out, and why is it important for SEOs?

“Fan-out is really interesting. You see it within AI Mode, and in AI overviews as well.

Basically, it’s a way of anticipating what the next step of the journey is. In AI Mode, if you put in a query, and you are looking for ‘best places to visit in Paris’. It will give you the best places to visit in Paris, but it will also think about what the second step of that journey is.

If I'm looking for ‘best places to visit in Paris’, it might give me suggestions of restaurants as well, in case I want to eat somewhere. It's not going to just focus on the Eiffel Tower; it's going to anticipate what I’m trying to achieve. Am I trying to build an itinerary? It's looking at the next step in the journey.

People Also Ask is another way to see what the next steps are. When you put in a query and you look at the questions that are connected in People Also Ask, they usually have a logical structure, and you can see that there's a bit of a journey there.

If I'm looking at exploring ‘best SEO tools’, my next question might be related to pricing, for example.”

Can you figure out the order that these queries should be in to get the correct fan-out?

“It's not about the order of it. One of the things that we have done with PAA (which is still very prevalent) is that we literally treat it as FAQs. However, it's not about creating a page that has every single one of those questions written out in a list.

It's not about having an article that just answers the questions; it’s about the intent. The article doesn’t even need to have a single question in it. As long as it answers those questions in terms of intent, then you're sorted.

It's not about the order; it's about whether or not that page responds to the intent behind those queries. I use AlsoAsked, which is a great tool by Mark Williams-Cook, and it gives you the PAA questions on several levels. It's brilliant because it gives you a journey of how people are searching based on your initial query, and what the next step is.

However, I'm not going to just take those queries and create an FAQ with them.”

What is chunking, and why is it important?

“Chunking is the most technical one because chunking is actually something that has been taken from machine learning. It's not an SEO term, and it’s the one that makes the least sense to me when it comes to SEO.

The way that it's been understood is, if you have a paragraph that responds to a query and a specific intent, you need to chunk it out so each one of your paragraphs is basically an answer to a specific question or intent. You can see how that would happen if you have an FAQ blog post. That’s how it’s been represented.

Actually, in machine learning, it has nothing to do with that. It’s related to how many tokens they have. The chunks vary in size, and it’s not about the context and the understanding of that paragraph; it's something completely different.

To be fair, I only know the basics of chunking, but from the basics, I can already see that it's not going to help me create my content because every model has a different chunking mechanism. What are you going to do? Are you going to write one article for one model and another version for another model? It just doesn't fit in my head.

I could be completely wrong, though – maybe I'm going to eat my hat, and there's actually something that I'm missing there – but for now, in my mind, the SEO version of ‘chunking’ just doesn’t quite add up for me.”

Emina, what's the key takeaway from the tip you shared today?

“Don't get bogged down by the technology side of things. Think about your audience.

The technology is there to help you understand your audience and learn how to build a website and manage your digital marketing channels in a way that your audience will understand and connect to, which will ultimately result in revenue.”

Emina Demiri-Watson is Head of Digital Marketing at Vixen Digital. Find out more over at VixenDigital.com.