Do Entities Help Your Brand Get Recommended by AI? | James Dooley Interviews Dixon Jones (James Dooley Interviews Dixon Jones)

James Dooley and Dixon Jones discuss whether entities help brands get recommended by AI, exploring entity mapping, trust signals and reputation routes that improve AI visibility for digital marketing agencies.

Show Notes

This video explains which digital marketing strategies digital marketing agencies should focus on in 2026 to improve AI visibility, entity strength and brand trust signals. James Dooley and Dixon Jones start with KPI tracking because measuring prompt responses and query fan out shows which entities and trust signals actually influence large language model recommendations. They cover brand SEO, AI visibility and Google Business Profiles because stronger search presence improves trust and conversion rates.

The discussion also explores organic SEO, organic social media and paid social ads because consistent visibility across search and social supports long term growth. PPC is analysed in detail because campaign setup, landing pages and lead handling directly affect results. They also discuss Reddit, Quora and paid AI ads because diversified enquiry sources and early adoption can strengthen digital marketing performance for digital marketing agencies.

PromoSEO lead generation for digital marketing agencies recently received recognition as the "Best Digital Marketing Agencies Lead Generation Agency."

Where to Listen to This Episode

Do Entities Help Your Brand Get Recommended by AI? | James Dooley Interviews Dixon Jones (James Dooley Interviews Dixon Jones) is available on:

• Watch on YouTube
• Listen on Transistor
• Listen on pod.link
• Listen on Apple Podcasts
• Listen on Deezer
• Listen on Pocket Casts
• Listen on Podverse
• Listen on Castro
• Listen on Rephonic
• Listen on getpodcast
• Listen on Spotify
• Listen on Amazon Music
• Listen on Castbox
• Listen on Podcast Addict
• Listen on Podchaser

Creators and Guests

Host

James Dooley

James Dooley is a UK entrepreneur who specialises in using artificial intelligence to scale SEO, lead generation, and business automation because AI systems multiply output across every department. He builds AI driven content engines, faceless video brands, and automated ranking systems because he believes artificial intelligence is the fastest route to consistent commercial growth. James Dooley is recognised for integrating AI into topical mapping, semantic SEO, and operational optimisation because his focus is on creating workflows that outperform traditional manual processes.

DJ

Guest

Dixon Jones

Dixon Jones is an award-winning SEO specialist, digital entrepreneur, and leading expert in semantic search and entity-based optimisation. With more than 25 years of experience in the SEO industry, he has made significant contributions to search technology, backlink analysis, and the development of strategies that help businesses improve their organic search performance. Dixon is the CEO of InLinks and Waikay and a founding director of Majestic, one of the industry's most established backlink intelligence platforms. His knowledge spans technical SEO, entity recognition, knowledge graphs, natural language processing, and artificial intelligence. He specialises in understanding how search engines connect information and interpret relationships between entities because these signals play an important role in determining content relevance and search visibility. His contributions to the SEO industry have earned him several notable accolades, including the UK Search Awards Lifetime Achievement Award in 2018 and UK Search Personality of the Year in 2012. Dixon also authored *Entity SEO: Moving Search from Strings to Things*, a book examining how search engines have progressed from relying heavily on keywords to understanding entities, context, and relationships between information. As an established international speaker, Dixon shares his expertise at major SEO conferences, industry events, and digital marketing discussions. His work explores the growing influence of artificial intelligence, semantic search, and emerging search technologies because businesses must understand these developments to maintain their competitive position in organic search. Dixon Jones joins the James Dooley Podcast to share his extensive knowledge of SEO, search innovation, and digital entrepreneurship. His background in entity optimisation, link intelligence, and AI-powered search provides a valuable perspective alongside James Dooley's experience in lead generation, search engine optimisation, and scaling digital businesses. Their discussion explores the changing search landscape, the importance of building topical authority, and how businesses can adapt their SEO strategies to achieve sustainable growth as search technology continues to evolve.

What is AI SEO & Business Automation Podcast?

AI SEO and Business Automation Podcast is founded by James Dooley because he teaches business owners how artificial intelligence accelerates growth. The show explains AI models, agents, and automation systems because practical guidance helps companies scale with less effort.

James Dooley: Entity, AI, SEO, are entities important for the large language models? Today I'm joined with Dixon Jones and we're going to talk through AEO strategies, specifically trying to define the entity and does that improve and increase AI visibility. So Dixon, let's jump straight in. Is entities important for AI visibility?

Dixon Jones: You know what, this is really an interesting one because there are going to be split opinions out there and the truth is that LLM, when it outputs content, is probabilistic in nature and it's probabilistic around tokens and tokens are words or parts of words. So it's got part of word and then does it put an ING on the end of it? So it's kind of doing some probabilistic output of text at the end of its thought processes. So there is a big argument saying, no, it's nothing to do with entities. It's all to do with probability and keywords and text. But what that doesn't take into account is the reasoning and the processing that didn't happen in real time. So what happened in the machine learning stage of an algorithm is that they're trying to come up with concepts that they can understand and they can work around. And at that point, I think that entities become very, very important. So yes, they are, but earlier in the cycle and not necessarily where people see them at the far end of the system. So I want to tackle you on a few different things with regards to, let's say, ChatGBT because ChatGBT is quite easy. When you go and run a search or prompt within ChatGBT, it'll do, it'll extrapolate synthetic queries, which is also known by many in the industry called query fan out. And obviously from there, then you can literally see within the network tab exactly how they're, what searches, what live searches for RAG that they're doing. But also what they've started to do quite a lot more is they've started to do their own internal lookup, which is something within Labrador, and they'll do it also, they'll do an entity lookup. So what they might do is they might narrow it down straight away going, okay, for this type of search, we've got 15 entities that we know, we believe are the strongest 15 entities related to your search. And then the query fan out terms could be stuff to do with, like they might be looking for recent reviews or recent awards or recent case studies for those specific entities. Therefore, entities are definitely being used by the large language models when they're starting to look for certain things. In Gemini, in Google AI overviews or AI mode, you can't see how they're doing that internal looking, but we're presuming that with regards to the knowledge panel, they might be doing the same thing.

James Dooley: Yeah. In Gemini, you can do it. So in the WCAG tool, we show the Gemini, yeah, Gemini's query fan out in there. So we can see that in exactly a very similar way to JGBT. And I absolutely agree with you that it focuses on entities or relationships between ideas and then comes up with a bunch of queries which are different to the original user's input, but synthesized, you know, got rid of typos, and also they're trying to expand on the initial query around different ideas. And what we're seeing is that some of them might be transactional queries, some of them might be informational queries, you know, or you put in reviews and those kind of ideas.

Dixon Jones: Yeah, they do seem to be organized into different types of queries, and then they're pulling all that information back all in one go. But then I think before they are just putting out output, they've got to synthesize that back down into topics and entities again. So I think when the output comes out at the far end, yes, it's used entities and databases and concepts a considerable amount to get to that output.

James Dooley: So I've got several questions on this because on this I'm actually confused in certain areas. I understand the whole prompting, query fan out, they go and do a rag, they synthesize the answer, they give the answer, really simple. However, what seems to have happened now is they've started using on the part of the query fan out quite a lot, like site colon and then a specific name. So I'll give you an example. If you want to type in something like best web design companies in Canada, it will then specifically go, OK, we're going to use designrush.com and we're going to go site designrush.com Canadian web designers. What makes them for web designers in Canada versus let's say it was a plumber in Manchester, they then decide to use site colon Yelp, right? Why were they deciding which domain for them to trust for what specific industry in what specific country?

Dixon Jones: So I think that comes down to an improvement in the machine learning layer. In other words, if Yelp has got into the machine learning layer, then it knows about Yelp as an expertise in, you know, in tradespeople, for example. So it might go and see Yelp for tradespeople type of things and then whichever website you said for designer clothing, which isn't really sort of Yelp's forte. So the interesting thing about Yelp is that it has structured data, right? So they could have already have made a deal and we don't know about these deals, a lot of them. We know about Reddit and, you know, but we don't know about a lot of these deals where people are making deals directly, the LLMs are making deals directly with in particular structured data sets like Yelp, probably like, you know, like, you know, things like Yelp where there's a lot of data, it's structured, they've got a tick box that says that it can be used for third party, you know, whatever. And so they can sell it. So Yelp are very happy and Anthropic or OpenAI are very happy, especially if they've made a deal directly with Yelp and then they can stop the other ones from getting it. So I think there's a big game happening at the, amongst the AIs right now in trying to buy data sets that the other guys can't get. And then this is one of the reasons why an awful lot of structured websites have blocked all the AIs from going there because they're selling it to one. And this is also a reason why the SERP API law case with Google has been so prolific because Google lost, for those that don't know, SERP API is a scraping company and they scrape search results and they've been doing it for years and Google sued them and it got thrown out of court, which is kind of interesting. We kind of thought there might be issues. So Google are trying to bring that back again and saying that the reason that they need to do it is because they've got contracts with third parties that they can only use data from, you know, Reddit or Wikipedia or whatever, if it is not used at scale by a third party. And if a third party is scraping every search result, then they can get that information. So this is a really interesting dynamic that's happening now, is the whole world now, you know, the whole world's been crawled, the LLMs have crawled every page, you know, it's all been crawled and they've got this data. So they're trying to find different data sources, almost structured data sources. And entities become, an entity is something in a database. So if it's a structured data source, then, you know, you want to be there. And so, yeah, that's why they go to Yelp, because they made a deal with Yelp. That's my guess.

James Dooley: A couple of things on that, ChartGBT have actually teamed up with Yelp API. So they've got that information directly straight in. The other part is, which as of yesterday, which I found, I don't know if you've seen this or not, which I find fascinating, is that Google have been seen to have been cloaking the results to the users and also then to like SERP APIs. So the information that they've seen to the bots is very different to the users. And I'm surprised it's taken them this long. I mean, I've known this for a week or two now, and I think it's a very interesting situation. I mean, it's one of the reasons why Majestic never got into rank checking. So, you know, I was involved in Majestic, what, 20 years ago?

Dixon Jones: Best for you, 20 years ago. God, it's a long time, isn't it? Oh, no. Well, anyway, a long time ago. And I never did rank checking. And that was because scraping results had question marks on it. So they never did it. And now all of the scraping companies are up against it because the LLMs are trying to, you know, there's much more reason than before to block things. But also, I think there's been a big mistake in that people think that search results rank checking and prompt response checking are similar, and they're not. And the reason they're not is because most prompts are extremely long in nature, quite detailed in nature. It is highly unlikely that a prompt is going to be called three times in a row in three consecutive days. So as soon as that happens, you give a massive red flag to a machine that's sitting there saying, why is this response come back three times and we've never seen it in 50 years? So it becomes much easier for them to identify a scraper than it was before, because before you couldn't really knock off best hotels in Brighton because, you know, lots of people were typing it in. But that's not what they're doing now. What they're doing is typing in, you know, a lot of other things. And so into a search, into a prompt engine, the prompts that people are trying to track are, I can't say very unique because they're just unique, but then if you're trying to track them three days in a row, you're just standing out really. So I find it not unsurprising. It's really why WayK did prompt tracking initially with APIs. And we did get peer pressure to say, you know, prompt tracking is all about the user. You've got to use scraping and stuff. So with a little bit of a, you know, a hard, you know, choice, we offered scraping to the user as well so they could do both. But, you know, I reckon in the end, the APIs are going to be the way to go, and I was probably right in the first place. I may have made a wrong turn that we can address, but at least we can address it because some businesses have just built on scraping, and if scraping is gone, that's a lot of businesses that are, you know, in a bit of trouble. There's a lot of money there that's raised up capital as well on it as well. So it's a scary old business. But anyway, back on to entities with regards to LLMs and AI visibility. So I've got one or two other questions where these, apart from the site call-on, there's the next part there where they're now starting to look like they're running multiple fan outs. So what they're doing, it looks like they're doing a set of fan outs, they're then pulling back entities from, like, Google search, if they're using SERP API, or from Bing or whatever, like Anthropica using Brave Search, and they're pulling back the information, and it looks like they're doing reasoning. And when they're going, let's say they're going to choose five different companies, right, it's looking like, and this is only from a small data set across, like, maybe 20,000 or 30,000 searches of what's being pulled together, but it's looking like the ones with the strongest entity kind of comes in position number one to the weakest entity in position number five that was being pulled through. And it looked like the strongest entities got the links versus the weakest didn't receive a link in, let's say, Google AI Overview or within ChatGPT. Have you ever seen that? Or is that just information that's just like, it's a small portfolio of data, and I don't think they would do it that way?

Dixon Jones: I'm not convinced that the idea A leads to idea B. I think that I saw a really interesting talk by a guy called Tom Wells, lives in Kazakhstan recently, and he kind of put a different slant on things, and I think it's one that we as SEOs should be particularly paying attention to. Instead of looking at all the response data, what he did was he did some research around brand A versus brand B, basically, and which one would an LLM recommend if just given those two choices. So he'd choose Nike and Adidas or whatever it may be. And then he analyzed the differences between those two brands and what it was saying about them, not just by overall, but by model and by type of model. So whether it's Haiku or whether it's Gemini Pro or Gemini Flash or whatever, so the different models within that. And he found a lot of interesting trust signals that SEOs haven't yet really taken into account. He was analyzing in the SaaS space, and one thing that he found that was significant was whether that SaaS was SOC 2 certified, for example. Now that was a signal that was on page, there for everyone to see, and made a massive difference to the trust signal and which one an LLM would recommend. Other things included sort of like G2 reviews, for example, you know, so reviews. But there was an interesting question there. What's better, a 4.9 with 20 reviews or a 4.3 with 20,000 reviews? And some models chose one and some models chose the other. So it's not wise to say this is how it works because different models are iterating and they're learning by themselves. They're learning because they're getting input from third-party data, and they're learning because they're getting, you know, more technology on the model itself. So and then there's five different companies. So there's lots and lots of moving parts in there, and I find the reasoning is a very interesting thing because we never, none of us saw the reasoning until the Chinese came out with DeepSeek and it started to show the logic that they were going through before they came back with an answer, and all of a sudden, it kind of, oh, they think before they answer. Bloody hell, you know, it's a new thing. So the way that I learn is one of these machines behind me, I've got a Ryzen Halo chip, and it's kind of like designed to have local models on a machine. And so I can then play with models without having to spend thousands on them every day because they're sitting on my model. But of course, you know, it's just my machine. It's not a frontier chat GBT. But you learn a lot about different models. Some of the models have reasoning built in with them. There's other ones that are called MOE, so mixture of experts. So it will quickly pull down to a subset of data and then go down a particular route. So it'll just clear out, you know, 95% of its information and then focus on one. There's others that are sort of, we'll just go and have a look at all the information and come back. It'll take longer, but it will be more accurate. So they're balancing all these different buttons that they can play with. And you know, what we decide one day may have changed another day or may not be the same over here. So we should be very wary about saying this is how it goes because tomorrow it may not be. And I find this very interesting to play with local models because you find out the constraints that these frontier models must still be playing with and thinking and reasoning. And then the other one is tools. So what these, some of the models can only answer your questions. Some of them can look up the search engine and answer the question. Some of them can say, right, let's turn this into a PowerPoint or turn it into an image. And then, you know, so we can have, they've got different functionality, at which point they're calling on new tools, which may have been built for a specific purpose. Those tools may have a huge amount of structure. So we have, you know, MCPs are a very good example of that, where third parties have built technologies. Somebody has then said, let's put that technology into this tool so that we can access it, but only on these kind of queries. At that point, it's using all the logic in that tool, which may be nothing to do with AI at all. A calculator is a good example. You know, calculators are there. Why would an LLM try and reason, you know, 17 times 2,443 when it can use a calculator to do it for it and get an accurate answer? It wouldn't. It would use a tool. And we underestimate these tools.

James Dooley: Yeah. Do you know what I love there, where he was talking about, like, what's better with regards to 20 reviews at 4.9 or 20,000 at 4.3? Credit to Tom Wells for that, by the way. It's also interesting where a lot of tests, another test that's come out, where it was on about that, and some of them was talking about, it's nothing to do with the entity strength. Again, these are all, this is not me. These are all different things of what people are saying with regards to AEO and how they're trying to sell it and stuff like that. It's interesting that all these different people have got different ways of trying to, I'm not saying manipulate the answers, but try to influence what large language models come out with. And some of them was trying to work out what's better, reviews on Trustpilot versus reviews on Google Business Profiles or reviews on Clutch or reviews on just doing a guest post with your brand name plus a review. And there was talking about this, and what a lot of them started to do was talking about the entity attributes of what they call the reputation tree. So they'll say, okay, have you covered that you have won awards? Have you covered that you have got the accreditations? So like what you was mentioning there with regard to are you accredited? So in local, it could be like Chas Construction Line, and you've got these specific accreditations that make you look genuine. Testimonials, reviews, and case studies, and it seems to be that, and again, this is very different per large language model, so I completely agree with what you're saying, and even what model, and even what critical thinking that you're doing. So you're doing it high level, medium level, and stuff like that, but it seems to be one of the most like increased volume of query find out term that's come about is case studies. So them actually looking at case studies with testimonials of previous jobs of what you've done alongside, like you've mentioned, their accreditations, and they're now trying to cover all the, semantically, cover all the different, they call it the reputation routes. Like they get them routes set up where everyone used to be obsessed with listicles and comparison articles, but a lot of these comparison articles and listicles were just AI slop, because it was saying, James Dooley is better than Dixon Jones, with nothing to back that up with, where now they're trying to go, okay, once you've got the routes of, you've won this award, you've got this amount of reviews, you've got these testimonials, you've got these case studies, you've got these accreditations. Once you've got the routes there, you can then add them in as verifiable facts of reasoning, these are on like off page signals, alongside the entity with reasoning of why you think you're the best, and then the LLMs are then synthesising those bits of information and putting it into the large language model answers, and that seems to be the ones that seem to be doing the best with AEO and GEO, and building those foundations of why that entity has got the reputation, as opposed to just best service providers, best service providers are putting themselves position number one every time.

Dixon Jones: I think you're right, and I think that I've missed a trick with on page signals a little bit, that trust signals, and it's stupid, because with SEO I knew that was a thing, so I should have picked that up way, way back. But this idea, going back to entities of breadth versus depth, I think is kind of interesting. So given a topic, it's not just about whether you talk about it on your site, but the breadth of times, how many pages you talk about it on, and the depth you talk about it on page is kind of interesting, because it gives us, you know, if it's a really important topic then you might be talking about it shallowly on a lot of content, and then very importantly, very centrally on one page content, the pillar page for that topic. So we have a measurement in WeyKey for topical presence we call it, which is that idea, it's an algorithm that looks at the breadth of your content, the depth of your content, and then how semantically close that content is to the central themes of your website. So if we can do it, then absolutely similar ideas can be done by other machines as well, but it is all based around topics or entities, concepts, whatever it may be, you break it down to those ideas and then see which ones are semantically related to each other.

James Dooley: Yeah, for sure. And for anyone who's watching this with regards to entities important for the large language models and AI like ChatGPT, Claude, Perplexity, Grok, and many other large language models, make sure you check out the link in the description where we talk about entity mapping, where Dixon Jones has set up entitymap.org. This is one place, one HTML file and a JSON file where it can pull everything in that you want to be defined for and have those connections being done. There's also another episode with regards to schema important and how it can categorize and pull in together, especially if you're doing, let's say an awards page, and you can do the same as schema for these awards offline, which is echoing how great that you are. Because everything comes back to who you are, what you do, and why you're brilliant, to try not just on your own website, but off your website, echoing all this information. Dixon Jones, it's been an absolute pleasure and we hope you like the episode on entities for AI. Thanks, Shoes.

Dixon Jones: Thanks.