Develop an interconnected data layer to show up and be understood by AI

Martha says: “SEO is now about managing your data layer for AI.”

What is your data layer for AI?

“We really have to think about the data we're putting into the machine.

Most people still think of schema markup as just trying to get visibility, but it's way more than that. By doing semantic schema markup, you're building a knowledge graph, which can be a contextual thing that allows machines to understand.

SEOs need to think beyond doing it for search visibility and start seeing it as the input we can use to control brand and understanding for large language models.”

Does this need to be done for every page on your website, or are you only talking about specific pages?

“We think of the website more as a system now. Websites are elements of the content that you're putting into the machine, but you want to use schema markup to define the relationships between those pieces of content and between different entities or topics on the site. It’s that connectedness that becomes really powerful.

You can't just think about optimizing a page; you need to think of that much bigger connected data layer, which I refer to as a knowledge graph.”

Are you trying to connect the data layer of your brand by appearing on other websites, or are you solely talking about your own website here?

“Backlinks, or interlinking within your website, are how we would have talked about it previously in the SEO world. Now, we want to think about how we're connecting those dots.

The important thing here is context. When we're interlinking between our sites, or even using backlinks to other sites or referencing things, we're not providing the context – and that context is the relationship between these elements.

To give an example related to schema, it might be that the page is about X product and you’re bringing in the context that X product is offered by Y brand, it has Z other products related to it, or that it solves a specific problem – and you want to define that problem, that industry area, or that category.

You want to provide specifics around not just the fact that it's connected, but the relationships and context within that connection and why the connection is important. Large language models need to understand those relationships to make accurate inferences.

There's a tonne of research around knowledge graphs. Gartner recently released their ‘AI Hype Cycle’, where all the generative AI things have come down and are now coming back up. One of the technologies that we need to be thinking about for this is knowledge graphs, because of the power that they have in helping large language models with their biggest problems.”

If you deliver more consistent and correct schema on your website and build a better knowledge graph, will LLMs be more likely to trust you?

“That's right. We've actually seen it solve very specific hallucinations, and it will also help solve brand understanding.

It's still going to help with visibility for pieces on the SEO side. For content teams, you'll also be able to have better coverage of entities and topics. Then, the third piece is brand understanding/brand authority.”

When you onboard a client, what areas of their website are more likely to be misunderstood by search engines?

“We’re finding that, often, it’s when there's something else on the web talking about it.

For one client, for example, the model was talking about a location being closed, even though it wasn’t. It was referring to an article about a time when they had done a repair in the office, but for some reason, that website was getting higher authority than the brand itself.

That was an example where they hadn't optimized or put schema markup on that location page. Once they did, and connected it directly to the brand, that became the authority for that answer and that query.

That’s a case where other things exist that provide a different answer – not owned by the brand but connected to the brand – which the LLM is looking at. In this case, it was coming from a news site versus the information from the brand itself.”

Why do LLMs love knowledge graphs?

“It comes back to being able to do accurate inferencing.

If you have seen me speak, you may have heard me refer to my connection to Kevin Bacon. I had a car that was in a movie directed by Kevin Bacon, and he used to drive my car. With that information, you can win the Kevin Bacon game because those relationships are explicitly understood.

LLMs have started using that logic, where they will jump along the data points that they have. The same is true with knowledge graphs. Knowledge graphs just allow contextual relationships to be defined in the data. That's why it has confidence in the data and in that inference.

We could probably do a whole other show on how LLMs use vectors of a site, but the vectors aren't accurate. They aren't grounded in truth, data, and relationships. That's why it's so different.”

How does a brand learn what information about them is contained within a knowledge graph?

“I like to think more about what it is known for.

Everyone's talked about entity SEO. A lot of marketers think of them as topics, or ‘things’. What's really interesting is when you do schema markup, and you start adding entity relationships to them.

We often talk about external entities and internal entities. External entities would be where you're defining something: a Wikipedia or Google Knowledge Graph reference. An internal entity would be the thing you’re talking about on your own site. It might be your products, your services, your people, or pillar pages that are talking about something that you’re experts in.

When you put that in the schema markup, you can start asking the knowledge graph questions. You can say, ‘What are the topics or entities that I speak about a lot?’ Then, you can start making inferences or asking questions around the topic. If you know the topic and the related entities, you can start asking, ‘What topics am I an authority on?’

In this world where there's a murkiness – a ‘dark funnel’, where there's lots of things we can't measure – the thing that brands can control is what they're putting into the machine, and what they're putting in the schema.

Are you talking about the right topics and entities? Is your content actually talking about those things? That way, you can figure out what other content you need to build and put in.

It's about understanding what you’re talking about in your graph – and you can also query graphs. At Schema App, we're about to launch our Model Context Protocol (MCP), so that people can query their graph and connect it to other AI initiatives.

SPARQL queries are a way that people would historically carry out a query in the knowledge graph or semantic technology community. Now, MCP, and some of these new AI standards for querying information or connecting agents to information, will make it more accessible and more powerful – whether that be internal initiatives or external agents, as we see those evolve.”

Can brands also measure what they get out of the machine, in terms of brand exposure and being surfaced within LLM search results?

“There are things we can measure today and things we can't. Google does a good job of showing us the impressions, clicks, and AIO citations, so those are relatively easy to measure. The harder pieces right now are with ChatGPT.

There are lots of new companies coming out that are trying to measure brand authority with regard to prompts. I think it's a directional measure rather than an actual measure, since the prompts that they're using to pull those numbers are typically way shorter than what we're seeing as the standard.

I don't know about you, but I write super long prompts with lots of context and lots of questions, and I ask for clarification before I get it to provide me with research.

Part of it is that we're in a wait-and-see mode. The best measure is conversions on your website and traffic, mostly because you can see it. However, you're not going to see some of the interactions and answers that are happening in areas that are providing services.

There are so many different AI chatbots now, and I think we're going to see them verticalized to where we won't see that initial interaction; we'll just see engagement lower in the funnel.”

How do you determine what type of content to publish, in terms of what’s best for different queries and where they are going to be displayed?

“It's like the old world of keywords, but thinking broader than that and considering the entities. Also, looking at the data you do have around how you're performing on AIO, which is still in Google Search Console, and tie that back to queries.

I would go back to first principles. Try to figure out (with that lack of information from ChatGPT on what’s actually performing) what you need to drive for the business, regarding what you need to convert on and be known for.

It goes back to brand strategy and even content strategy. What are the entities or topics you really want to be known for? Make sure you have coverage there. Then, as these new metrics come up for brand citations and AIO citations, which give us more specificity, it will allow us to know more.

The piece that we're working on a lot with our enterprise is how the brand is actually talking and whether it has the content that allows them to be an authority on the things that they want to show up for.

When we finally get accurate measures from the ChatGPTs, it'll give us more continuity on interactions or what topics people are actually asking about, which will then allow us to have grounded data to tie it back to.

Right now, I'm encouraging brands to just go back to basics. Look at what it is that your customers are looking for and what you’re trying to sell, and then make sure you have great coverage on that.”

How do you ensure that you’re covering all the bases that your customers might be looking for while remaining relevant to your brand and your audience?

“It's an iterative approach, frankly. Most business owners think, ‘What are people engaging with today, and what do we want them to engage with?’

When you see sales in certain areas, you go back to basics. Ask customers who have bought today why they bought, what problems they're solving, etc. It really is this beautiful thing where marketing, content, and SEO are all working hand in hand.

I look at whether the data that’s being derived from the content is talking about the things that customers are looking for. We've seen elevated engagement and cross-functional work across brands where SEOs – because they have the data, they can look at entity coverage, and they can look at what parts of the sites are actually optimized – can say, ‘We have big gaps in this play.’

Content can then use that data and build that data. Then that goes back to marketing, and whether they are seeing the leads come in on the areas that they want.

We're seeing more cross-functional work than ever as these things come together, driven by AI coming out, which I think is really cool.”

Should SEO be leading this, and what does an internal SEO 101 training session look like nowadays?

“I heard one of my customers say, ‘I'm so glad that I now have the language to talk about entity coverage and entity authority, so I can go to content teams and have data to support the recommendations that I’m making to the business.’

SEOs are the heroes now. They can translate what may have been nerdy technical SEO pieces and educate about what needs to be in the data, how the data is derived from content, what we need to build into the content, and how that supports the overall strategy.

It's connecting those dots. If SEOs can make those leaps and explain it in those terms, it will really give them more of a seat at the table than ever.”

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

“Don't just think of schema markup for visibility.

Think of it as your data layer, and a way to engage across your enterprise to make a difference in how you show up and are understood by AI.”

Martha Van Berkel is CEO at Schema App. Find out more over at SchemaApp.com.