Latest / Semantic SEO Podcast / Entity Mapping Explained: How and Why You Should Build One | James Dooley Interviews Dixon Jones (James Dooley Interviews Dixon Jones)
Transcript
- James Dooley: Entity mapping. How and why you should build out an entity map for your website. Today I'm joined with Dixon Jones and we're going to be talking about, we're going to go deep and there's going to be multiple episodes talking about entity SEO. So make sure you check out all the links in the description. But Dixon, for anyone that doesn't know, what is entity mapping? Dixon Jones: So entity mapping, well, it's pretty new actually. Well, if you go to entitymap.org that's new. But the idea is that the world, your business doesn't think, you know, the world doesn't see your business necessarily as a set of keywords and things. What they see is, and what a machine will see, is the topics that align with your business. So if you're Nike, then trainers are really important. You know, maybe hats and sportswear and sports people are really important. But if you're, you know, Ford, then Mustangs are important. Now the question is, is a Mustang a horse or is it a car or is it an airplane? So the world is organized and our human brains and machine brains are organized in topics and concepts. So entity mapping is really the process of trying to decide how your business relates to ideas that people know about in the world. So people may not know your brand, but if you're the only brand that does blue widgets in Hong Kong with, you know, 24-hour delivery, then that's the things that are associated with your brand. So entity mapping is a file, it's actually two files if you do it through entitymap.org, which is an open standard anybody can use, no need to log on, no, you know, it's all just there for everybody. It's a little bit like, you know, it's llm.txt or cats.txt or, you know, grounding page or site map on steroids really. What it does is it creates a couple of files which tells machines how the topics that you talk about on your website relate to each other. And that's important because at a page level, all the other tools and ideas around this, site map and stuff like that, talk about what your pages are, where your pages are, but you're relying on a machine to connect all the dots between the conversations that are on those pages and or a human being. And an entity map forces you to think business-wise about what the relationship is between this project and this audience and this feature and etc. And it shows relationships between ideas. So it says you can't do this until you've already got this kind of thing. So that's a whole new level of mapping. All things up to now have been about like schema about on page or schema about your organization, but not really the relationship between the different topics and ideas you have on your page. So entity map creates two files, one at entity map.html and one at entity map.json on your root drive. And literally both of them do the same thing in different ways, but they will tell one reader of that page everything that's important about your business and how it connects to the other ideas in your business. James Dooley: Yeah, absolutely makes sense. So my next question on there from there then is why does entity map.org create two files? Why is it not to create one? You mentioned that it does a HTML file and it also does a JSON file. Why does it do the two? Dixon Jones: Excellent question. The truth of the matter is that we all know that HTML files get happily indexed by Google and Bing and everything else. And that's what they do. They just soak up HTML files. So you don't need any new, you know, we don't need to have any technology created by a search engine or by Anthropic or Open AI or anything like that for them to digest that information. They can all digest that and see it perfectly well. And it's great. And it shows all the relationships and it shows where on your website you talked about this idea as well. And everything's all in one place. The thing is that that is very structured in the way that it's built. And it's built around a JSON data set anyway. So providing a JSON file allows machines to actually save a lot of energy and a lot of effort in understanding it if they wanted to. So I'm not saying that we've got evidence that people are using the JSON file at this particular point in time. But certainly if you feed that JSON file to Claude and you feed the HTML file to Claude and say which one's easier for you to digest, it says the JSON one is easier. Now, it's not necessary that at this stage they're going to look at it, which is why we've got both. So that we know that there's technology for now and there's an offer for machines to take a cheaper route in the future. James Dooley: For anyone that doesn't understand the difference between an HTML and a JSON file and why the JSON file is easier for, let's say, Anthropic or Chat GPT to dissect that information, can you briefly understand why? And can both the HTML and the JSON file both index within Google Search or not? Dixon Jones: So JSON files you don't generally see in Google's index. So you won't see a JSON file in a Google index. Which is a really good reason to have the HTML file. But the HTML file can literally, not literally actually, but a lot of HTML files, especially data-driven ones, will call a JSON file and populate all this information from a JSON file. So if you think about JSON file as being just a structured database on a page, I think that's a good way to see it. So it's almost like a spreadsheet written for machines. And they're used all over the place by developers because they're very efficient ways of communicating data. And so it just means that they don't have to build a browser, et cetera. They can literally take that JSON file and feed it into a program that they might use. So it's almost a more universal way of communicating information. It's generally a much more structured way, even though the HTML file and the JSON file are exactly the same structure. They say the same thing. It just takes a lot more energy to read HTML and then effectively convert it back to JSON. So it would be better if machines read the JSON file. They don't yet. Well, they don't in the typical course of things, but they can. They're completely capable of doing that if you give them the JSON file. And then with regards to if someone comes and uses EntityMap.org and they get the two files, what type of URL structure are you putting this on? So let's say you've got the brand name, JamesDooley.com. Forward slash HTML. So forward slash EntityMap.html and forward slash EntityMap.json. And then typically the protocol recommends that you then link to the EntityMap.html in a footer in the same place that you'd put a sitemap or an index or about us or whatever so that it can get indexed reasonably by machines. And then the HTML actually refers to the JSON as well. So the HTML then defines that this is the specification at EntityMap.org and says this is the specification that if the AI wants to go and look at it, they can. And then also goes through and shows where the JSON file is if they wanted to go and see that instead. And then goes and explains all the stuff on the page, which is really structurally easy to understand. It's easier to build one than you think as well. You can just literally go to Claude and say please take these top 20, 30 pages on my website, the ones that are really important for your business. And then read the specification at EntityMap.org and write me EntityMap files. I mean there's better ways to do it, but that is a perfectly acceptable way to do it. And you just have to check the output because the LLM might have got it wrong. James Dooley: So with regards to that, just to clarify on that. In the footer, you only link to the HTML file, but from the HTML file it'll link then to the JSON file for the EntityMap. Dixon Jones: Absolutely, yeah. So the HTML file already refers to the JSON file. We understand that the JSON file isn't going to get indexed and put into a typical search result, but it can get used. It can get used already. For example, we've written an MCP, so a little skill that you put into Claude. So I have one on my Claude where every time it goes to see a website, it checks to see if there's an EntityMap file. And it checks to see if the JSON's there because it uses a lot less tokens to read the JSON file than the HTML file. So there are plenty of use cases that you can use for it right now, but you don't need to link to the JSON file. That would probably just confuse a browser because a browser isn't used to linking to a JSON file. James Dooley: Yeah, it makes sense. And then with regards to creating the EntityMap, should we be trying to get all hyper-specific data for the brand? So not just obviously what's on the website. Is he trying to get all defining the founders, the founding date, every last little bit of information with regards to who the entity is? Dixon Jones: In a way, that's a business question rather than a technical question. Our stance or our own business's stance is that you build a map that is going to be a good representation of your business. Now, it does have to have content on your website for it to be a version of your business. Because what we're trying to do here is we're trying to stop the LLM from hallucinating. And we're trying to stop the LLM from not citing your source if we can get it. So what the map does is it says, hey, this feature relates to this feature or this person relates to this feature or whatever it may be. And the relationship is belongs or requires or is excluded from or there's a whole load of 59 predicates, I think. So sort of almost verbs that allow you to explain how two things can relate to each other. And then it says, and here's the evidence. And it'll pull out a quote from your website and then provide a link back to that quote. So the idea here is that you're trying to reinforce your interpretation of what your website is saying about it. So now that may not be everybody else's interpretation of the world. So if you take I'm going to I'll take McDonald's, you know, so McDonald's had some bad press in the past about its, you know, about its food production techniques and stuff. And now it has a huge section on sustainability and other bits and pieces. And they put things in McDonald's. So they've got their view of reality and some public have a different view of reality. You want to portray your view of reality as best you can. That doesn't mean to say that you have to go down to every ingredient of every burger because doing that is detracting from the important parts of your business. So you need enough in there for the business to be well understood and well represented by an LLM in particular. But and a search engine as well, but not so much so that the important messages get diluted. James Dooley: And then with regards to the entity map dot html, what is if anyone is listening to this, they might be thinking and getting confused with this is like LLMs dot txt. What's the difference between LLMs dot txt and an entity map dot html? Dixon Jones: Well, firstly, it does get crawled and read. And so so it's it's early days for it. But in our tests, it became the third most looked at page in Google Search Console. Well, actually, Bing, we kind of looked at as well. So Bing was the first one that picked it up. And and and also they've got sort of a discovery click information, which is Google Search Console doesn't have at the moment. It's only got a discovery impression information. So it does get picked up. It does get read and does get understood. Part of that is because it's a dot html, not a dot txt. That's a big, big advantage, I would say over it. The second thing is that an LLM dot txt is is is not representing the relationship between between ideas and doesn't link back to the evidence that backs that up. So it's this whole relationship between ideas that totally lifts it up to to give an idea of of why we before we launched it, before we launched the free, the open standard. We had a call with a guy called Raminathan Guha. He is one of the co-founders of schema.org. He's worked at Google. He invented the RDF format. He's worked OpenAI. He's worked at Apple. And now he's a research fellow at Microsoft. And he was kind enough to give us a 60 minute time with him as we went through this and said that is this you know, are we are we being idiots here? And his quote, which I followed up by email afterwards to make sure I was allowed to say this is, you know, his quote is this is a good thing for the world. Now, he's not saying that Microsoft are going to take it on board and use it or Google are going to take on board and use it. He's saying that this is needed in in in the search engine world because we keep on creating content page by page. And OK, we can link those pages together. But then for a machine to understand the relationship between those concepts, it's got to read all of the pages and then come up to make a decision. If you can infer that decision in one file for the major concepts, it's going to reduce hallucinations. James Dooley: Yeah, for sure. And on there, you mentioned that it's the third most traffic page when you did it within Google and within Bing. This is on the example that we had. Dixon Jones: Yeah, we've we've we've got around about we've seen about 500 of them in the wild now. So they've taken off, but they're a long way from being proper. Obviously, with it being linked to in the photo, the Google bot and stuff's going to be crawling often because it's linked to in there. Does it do you think that the URL? I mean, I've seen some people doing like certain pages like LLM dot info or they've done certain ones where it could be like a info or some of them just put it as an SSL. T standing for a single source of truth about defining the brand. Does it matter what the URL is or would you have it as being entity map dot html? I think you've got a choice. I mean, I mean, technically you can do whatever the hell you want. I think the advantage of having a standard that everyone can can jump into or is is that it's it's being thought through. People can argue about the standard and we can update the standard over time, but it allows it. If if ever the LLMs want to say, right, we do need one of these. You know, what are we going to go with? You know, they don't want to just create a new one out of out the blue, really. So they might as well start looking for the ones that have got some semblance of sanity and some semblance of traction. So what I'd say is making up your own stand, your own standard is is fine. But I don't think it's going to do the same or your own, you know, dot html or whatever page is fine. But I don't think it's going to have the rigor and structure that is is really needed in entity mapping. So in the system that we've got, every entity has to have a Wikipedia or a Wikidata URL. So you can't call it an entity if it hasn't got one of those. That immediately screws up a lot of SEOs who want to put themselves in as an entity right on day one. We are thinking about making LinkedIn profiles acceptable. That's part of the part of the discussion process, which would allow pretty much anyone to have their own personal entity URL, if you see what I mean. But but going beyond that, you start to break things, you start to make up entities. And it's kind of like trying to create your own Wikipedia page when you've got no earthly right to do so. And we all do it. We're SEOs, you know, if we can, we will. But but but this is one reason why you don't you want to have some kind of structure into the system, because otherwise it's not structure at all. James Dooley: Yeah, for sure. Otherwise you just increase in disambiguation. You want to make sure it's mainly there for confidence and clarity. And if you start adding things in there that's not instilling confidence and clarity, then it defeats the object. One last question, because I love the idea of having an entity map. I also like the idea of entity maps with regards to it's easy to feed, let's say, anthropic or chat GPT or link builders or you're doing a press release and someone says like a digital PR company says, can you give me all the most important bits of information about who you are and what you do and what it is? You can just send them that one page. It's that single source of truth that's got all the information in there. But one or two people get worried about creating like the entity map HTML because they feel it could cannibalize with like the about page. What's your thoughts on that? Dixon Jones: So it's quite likely that it might link to your about page anyway. I think an about page typically will will say, you know, who the who the who the directors are, maybe your story and your main, you know, your main offices and those kind of things. So that does have structure in its own right. What it doesn't have is a relationship between your different products and and, you know, departments. So, you know, if you I mean, if you're if you're a national health service and you've got and you were trying to do one, then maybe what you would want to do is this surgery over in in Woking specialises in this kind of treatments and this one over here specialise in this kind of treatments, etc. So you might want to do something like that, for example. So I think that you you would have stuff that is quite dry unless the person unless unless a reader really wants to, you know, take that and then to create, create, store, create, create their own bio or whatever for the business. So so I think that they are different. However, there will be cannibalization. There will be the same text because it literally will take bits of text that are on your website and refer to those and give those as evidence of why this relationship relates to this, this relationship. In other words, there's there's evidence on your website. And what we're trying to do at that point is hope the hope that the AI, you know, trusts what's said on that page because you've got stuff on other pages that backs it up. I have to put the caveat in there that this could be, you know, people it's like meta keywords. I can just see people trying to say that their businesses can do absolutely everything and is going to, you know, make millions and millions and millions and just sign up here and they're just going to put crap in there that can't be justified. So, of course, like every other web page, an LLM and a machine is going to judge its credibility. I mean, I'm surprised that, you know, none of the LLM businesses, Anthropix and, you know, and OpenAI haven't bought out a Moz or a Majestic or a backlink database because the backlink database, particularly Majestic's anyway, can really do the same as PageRank used to do. And right now, all of the LLMs are having to rely on search results because they haven't got this backlink relationship stuff that works out, you know, whether a page is strong or not. And so I think but they will all do things like that. So, you know, I don't think you putting a page up there gives you necessarily credibility. There still needs to be credibility in the website for the machine to take that page seriously. James Dooley: Yeah, for sure. Anyone who's listened to this episode, we hope you like the episode about entity mapping. Have you got your own entity map put onto your website? If not, I strongly recommend going over to EntityMap.org. Go and get that created. Get that JSON kind of URL as well as the HTML. Make sure you link to it in the footer to the EntityMap.html page. Dixon Jones, it's been an absolute pleasure. Thanks for having me. Dixon Jones: Thanks.