Introduction

This is me, I'm Berisha Gamba, I've been working in the semantic AI -powered SEO space for quite some time now and I work at WordLift, these are some of our clients and I personally work with big enterprise clients and today we'll be focusing on a use case for SEO Luxottica and we cover most of their e -commerce markets and brands.

friends.

From Google-Only to a Multi-Player AI Search Landscape

So as Omar said, we are transitioning from a situation where Google was dominating AI search and the search industry. Still is, of course, but now we've got more players

in the field, like, of course, Bing and OpenAI, and now you can actually buy from this platform. So there's a big competition out there. And of course, you have to evolve as well in order to keep up.

1If you are a website owner, if you own an e -commerce, you have to make sure that your website can be read by the machines. And this is what we do at Wordlift and this is what I do daily.

The Core Challenges: Hallucinations and Fragmented Company Knowledge

So let's start with the issues. I prefer to start like straight to the point and then eventually delve on how we can solve them so models hallucinate right that's the the main problem there that's the main issue

and then there's the fact that company knowledge is often scattered sometimes bigger companies do not even know what data points are available to them and somehow this knowledge has to be modeled and engineered and of course companies do not always have a persistent source of truth They have many tools they rely on, and the information is a little bit here and there.

They provide you with any type of reporting, and then it's up to you, I mean up to us as an agency, to make sense of all of their data.

How do we tackle this? is.

The Solution: Building and Publishing a Knowledge Graph

We work with a technology that's called Knowledge Graph and it's a way to connect a lot of data points and to make sense of it. And we publish this Knowledge Graph online.

The Knowledge Graph contains either information about products, information about a specific company also just simple content can be structured inside a knowledge graph we

Agent WordLift: Automating SEO and Content Workflows

have built an agent agent word lift is our tool and it helps agencies and digital marketing teams to streamline their SEO tasks so not only SEO tasks tasks, also newsletters and writing content, so this is what we'll see today.

Use Case Demo: Choosing Sunglasses for a Round Face

I'm asking Agent WordLift to help me find a pair of sunglasses for my round face. What's the right pair of sunglasses for me? So I'm asking agent WordLift.

This is also a query that you can type onto any platform like ChatGPT, and what I want to show today is how building a knowledge graph can help reducing the number of clicks from when I start thinking about wanting a pair of glasses and the moment where I purchase them because this

is what we do we try to short the path from what trigger a user like I want to buy something and and the moment that they buy it. Because when you structure your knowledge, it's more immediate both to users

and to search engines on what to find. We start with the knowledge graph. Everything at WordLift starts with the knowledge graph, and it's fair that we do as well today.

Step 1 — Modeling the Website: Templates, Entities, and Structured Data

so Purcell .com is one of the website we work with. I'm asking, okay, start from my

website, identify the main page templates like the home page, category page, product page, my blog post, my FAQ pages. For each template, list the entities that I should should add to my graph.

Generating JSON-LD for Each Page Template

And then I'm also asking to generate the JSON -LD. So we are generating something that you can directly add on the header of your pages in order to have structured data on your pages. And DHS does that.

It provides with the, I didn't copy the whole thing because it was too long, but it provides with the templates templates for every, it can detect every page template, and provide me with a JSON that I can then download and directly ask my dev team to add to my pages on my website. And this is the knowledge graph structure.

So agent was able to help me do a task that requires several people in tech team to do and I'm asking for some analysis right right now I have organized the knowledge what do I do with it what's the next step so you may know that now

Step 2 — Query Fan-Out: Covering the Full Set of User Intents

we have a mode also in Europe and it's a way in which Google search breaks down query into dozens of multiple sub queries and the agent has a way of simulating this approach and providing you with the list of sub queries that you can then cover on your site in order to own your space in order to own for example we are still talking about the round face sunglasses so if I want my my website to rank for round face sunglasses, I should cover multiple sub -queries about that.

And I'm asking, take this query, expand it into 15 to 20 distinct user questions, and provide me with everything that I need to cover. And here we have the user questions, and also divided by clusters if you have different category different category pages that can accommodate more than one question. So we've got that.

Step 3 — Curating the Graph: Identifying Missing Entities

But now, I have built my knowledge graph. I want to curate it, to maintain it, I want to expand it. So based on the questions above, what are the entities, the additional that I should add back to my graph.

And so we are using these Purcell .com USA sunglasses round faces and provided me with the conceptual entities that are actually missing on that page. So what we are saying is your page is missing these top entities that could help having them on the page could help rank this page higher. in the on Google search so we've got everything that we need now what we are

Step 4 — Creating Content and Validating Quality

missing is content so now we are ready for content generation and I'm asking write me a paragraph that answer this specific query so this is one of the sub queries that I generated during the query fan out simulation and it provides me with the direct paragraph and it explains a little bit how I should write

it and add it to my page and then I can also ask okay but is this content any good like because as we know we shouldn't just copy and paste something directly from AI and this tool helps us in assessing content quality so I ask is this good?

Scoring Coverage, SEO Relevance, and Hallucination Risk

You should return the coverage score, like is this paragraph actually covering what I asked? Does this paragraph contains the entities that it should and hallucination, factuality concerns, missing entities from the knowledge graph and a short improvement checklist.

So we are asking a lot of things that are very useful. And we've got the replies, so the quality score is 62 .9, purpose score is 62, so the maximum is 100 a year of course. Then there's the SEO coverage score etc etc.

So this also goes to show that usually the first generated content is not that good even if I generated it fairly probably we need another iteration so usually a couple iteration produce a content that's like 70 to 80 as a content quality score and this is what we did for Purcell, like the whole process we did for Purcell US.

Outcomes: Rankings, Click Growth, and Visibility in AI Search

So if you search for Purcell sunglasses for round face, but even without Purcell, you should see this page on top.

And I mean, we help the business reaching the users that wanted a pair of sunglasses classes that looked good on their faces. And this goes to show what I'm saying is that

it does not mean that having a knowledge graph can ensure that, but it works as an information amplifier because we are delivering the information in a machine -readable format and makes it it more recognizable by by AI and these are the results of the personal

Search Console Impact: Clicks, Impressions, and Informational Queries

implementation this is the curve of clicks and impressions on the side like this is the before this is after and this is the curve that represents the the informational queries. So the how to, what are, and where.

Because having tailored content that covers the whole space of the round face sunglasses help them cover the majority of the informational queries that people are actually searching for.

Emerging Channel: Referral Traffic From LLMs

And this is the traffic on AI only on the LLMs. and we can see referral traffic over time showing an increase of course based on the past year when not many people were using LLMs but I just wanted to show that there's actual traffic coming from there and also like it's a news of

yesterday being just added the data from LLMs to their webmaster tools and I'm sure that Google will follow shortly.

Key Takeaways: Data-First SEO as AI Infrastructure

So very quickly a few takeaways. So if you are structuring in a data readable format your content and your data you will be more recognized by the LLMs.

Domain Focus and an Ontological Core

You should model your domain so stay in your domain do not write about something else you don't have to cover everything but but you can cover more thoroughly what you are doing well.

1We have to focus on defining our ontological core, the 30 to 50 core concept of your business, and just stick to that.

And then that visibility on AI is an infrastructure. It's not just mere SEO. If we start from data, if we are grounded in data, It's then, of course, easier to maintain such an infrastructure.

And Knowledge Graph can also help supporting not only content or SEO, but also sales and customer support. I mean, they are a great decision -making tool.

Closing Perspective: Building Systems That Empower People

And then I really love, like, Tim Berners -Lee is one of my biggest leaders that I really like. And he said that in the age of AI we have to build systems that empower individuals because at the end of the day we want to help humans do more and

Conclusion

That's it Thank you Thank you so much