Should Businesses Track What AI Knows About Their Brand? | James Dooley Interviews Dixon Jones (James Dooley Interviews Dixon Jones)

James Dooley: Wecay.io, what AI knows about you? Today I'm joined with Dixon Jones, the founder of Wecay.io. And I want to jump straight in Dixon and say, why is it that you created this brand?

Dixon Jones: So we, when we tried to come up with all sorts of names for the brand, and then somebody said, look, why don't we just call it what it is? You know, it's what AI knows about you. And then we shorted it to Wecay and we thought, great. Well, now no one can spell it. No one can pronounce it. And we don't have the .com, which wasn't necessarily very clever, but it does make people start with, you know, the acronym, which is good. So the reason that we created it was, it came out of our other product, InLinks, which is an SEO technology. But it came out and we did it as a separate thing because when LLMs came out, it was quite clear that there was going to be a market for understanding, you know, how to get your message, your brand message in the prompt responses of whatever it was that people are going to type into these machines. And it took a little bit of time. I'd say it took a good year for people to start understanding the idea of transformers and all the things that happen inside those LLM boxes. But we decided that we wanted to build a new tool for it because if we tacked it onto our other business, you know, we'd confuse people with two different lines of thought. So what we decided was that we had to help customers or help users understand what the AI actually knows about you as opposed to what comes back in the prompt response, which could be very, very different because there's this whole layer of search that goes on top of the LLM machine learning. And that disguises a multitude of problems. I mean, if LLMs were perfect, they wouldn't need a search engine, right? But they need a search engine almost every time. As soon as you start doing anything above how to boil an egg, then they're tending to look up on a search engine. So I don't think Google's dead yet because, you know, they've got the best search engine, so arguably. So I think we wanted to build technologies that would help users understand what AI knows about you and probably more important, what it should be knowing about you that it hasn't picked up with because that topic analysis, that topic gap analysis gives a strategic direction for a marketing department on what to do about it.

James Dooley: Yeah, for sure. And anyone who's listening to this, make sure you check out the link in the description where me and Dixon go through in depth about entity mapping and having an entity map put onto your website. My next question to you, Dixon, with regards to what AI knows about you, I want you to give me an example where someone wants to increase AI visibility, whether it's in ChatGPT or Cloud or Google AI mode, and they're wanting to know, okay, what bits of information do they know about the brand that might be incorrect? Is that something that within Wecay.io that they could be looking at to say, whoa, we need to fix this, we've got a problem here. They're pulling in the wrong information or they're hallucinating. Like, is this where Wecay.io can be helping people?

Dixon Jones: I've got an example of a friend of yours, actually, Jason Hennessey. So Hennessey Digital are a US agency. And way back in the day, oh, and if you asked, this was about 12 months ago, so it's probably been fixed now, but anyway, if you've asked about 12 months ago, what do you know about hennessey.com? Then not unsurprisingly, some of the LLM said it was a brandy company because he owns the EY and hennessey.com is just Y, I think. So that's not surprising, but it's an example of a hallucination. They put in his type, his brand, his URL, and it came back saying brandy. So there's misunderstandings in the LLM, all based around probability. Obviously, a lot of people type in the brandy, so we'll talk about the brandy, and let's talk about an SEO agency. So that's understandable. What was more interesting for me was that it also came back as it's an auto parts company. And this came back, by the way, Jason said I can talk about this when we found it. So I hope I'm not stepping on his toes too much. But when we found it, he went, wow. And what had happened was that back in the days, early days of SEO, there was a Hennessey auto parts company, and being the good link builder that Jason was, he went out and said, look, some people are gonna be coming to you and they want an SEO company. Why don't you link through saying, we're not an SEO company? And he put SEO company as the anchor text back in the day. And now, of course, a machine goes down, sees SEO company on Hennessey auto parts, and as a result, there's no links in their crawl because their crawl was not necessarily taking the link information at all, just takes the words. So it then associated the SEO company with an auto parts digital. This is the kind of example that happens a lot when you've got a name like Majestic, it's quite generic in nature, that can cause an issue. But you can also cause errors and hallucinations when you have a change of director or when you have a new product line or you have new things happening out. The machine has remembered things from a year ago or more and it just is embedded in its system. It doesn't really remember where it got this information from. And if your version of reality is different to their version of reality, you've got a problem because when you ask a question of an LLM or when a user asks a question of an LLM, it's gonna go in and it's gonna say, right, what do I know about that? Do I have some brands that I need to go and look up on search engines now? And if you're not in the brand in the right context, then you're gonna be brought out in the wrong context. So that's why it's important. Getting the accuracy of the underlying LLM right is probably more important than seeing whether you are in the results for what's the best name my product here.

James Dooley: Yeah, for sure. There's one that I'm probably showing my age here now. I remember you back in the Majestic SEO days where then you rebranded for Majestic SEO to just Majestic, but then you had the issue yourself with regards to there was a Majestic Wine. And for a while, Majestic Wine was top trump in Majestic SEO. What then, if someone sees that or looks to rebrand or they're seeing with regards to what AI knows about you is slightly different. What can people be doing to try to instill a stronger entity to make certain that AI does understand, okay, this is who we are, this is what we do, and this is why we are brilliant.

Dixon Jones: The solution depends on the nature of the problem. So the trick really is to find the underlying source that provided the wrong information. So in the Hennessy example, a conversation with the auto parts company might be a good thing to do. It could be as little as changing the anchor text really. But so we had a problem with our InLinks brand where somewhere someone said it was taken over by SEMrush. Now, we have not been taken over by SEMrush, it was bad. But what happened was we were able to track that back using our tools, but anyway, you could do it anyway. To the original source, to the webpage that had made that claim, it turned out that webpage had done a really quite nice write-up of the product, but they had put this, oh no, but they'd said that we were a rank checker. That's what it was. It wasn't that we'd been overtaken. It was that we were a rank checker and we don't do rank checking on InLinks. So we just reached out, pointed that out to them. They were so much easier than getting a link off someone. They were very keen to keep their data accurate and up to date, so they were happy to change it. And we haven't seen that in a response again. But there's a bunch of different things that could be going wrong. So what we do is we literally ask AI Claude, what do you know about this website? And all the rest is a little bit of magic, but whatever comes back, we turn that into one-liners, just again with another AI thing saying, turn this into one-line facts. And then we've got easy facts that we can confirm or deny. So we can just go down with the client, say, is this all right? Is this all right? Or flag the ones that we think are a little bit dubious. Anything that's dubious, we can click back to the original prompt that generated that false fact and see what happened. The things that may have happened, apart from the ones that we've said, is quite often it refers to pages that don't exist. So that's an interesting thing. And it's usually to do with, or in our experience, it's to do with navigational structures. So we had an example with Ikea, where they've got ikea.com slash, I can't remember the URL, so let's say chairs. And it didn't exist, but they did have slash kitchen slash office slash bedroom. So they had a structure and the AI had assumed that chairs had its own webpage. And it hadn't. That's an easy fix with a 301. So it's a case of, you've got to have the ammunition to find the problem and then hammer it out the system wherever you can. And that may be something on your own page. It may be something on someone else's page. It may be a technical issue that has arisen, but they're usually fixable. It's a lot more of a problem if the issue has become embedded in society, for example. So what I mean by that is, if the world has decided that, I'm trying to come up. Okay, so if you're British, you'll know this. If you're anybody else but British, you won't know this. And the ones that aren't British will get it right. So everybody that's a kid in Britain knows that carrots make you see in the dark, which is complete rubbish. It contains vitamin A. It turns out it was something that Churchill put out as propaganda in the war to explain why the pilots could see all the Luftwaffe. And it was put out on radio and also got the kids to eat carrots as well because that's all you could eat in the war. Complete rubbish. However, all Brits believe it. So we'll talk about it. So there's a much greater chance that we'll talk about it online. There's a much greater chance that if you ask an LLM, can carrots make you see in the dark, that it will say yes. So as a narrative becomes embedded in society, if it's a false narrative, it gets worse and worse and worse and harder to fix because there's multiple points of failure. So that's why you've got to fix it for your brand quick before it becomes out of hand.

James Dooley: And if someone's using Wecay.io and they're using it for AI brand monitoring, have you found that there's a big difference between what, let's say, ChatGPT knows about you versus what Gemini knows about you versus what Claude knows about you? Are you pulling all the different large language models and fixing it across all of them?

Dixon Jones: Yeah, so we have a completely different module for what AI knows about you and what AI says about you. So what AI says about you is a user typing in what's the best hotel in Brighton or whatever it may be. And it comes back with some, hopefully, some hotel ideas. But that side of things is not really so good to start tracking until you fix the underlying core issues. So, and I think that all the millions of, thousands of prompt tracking companies that have spun up over the last 12 months have gone straight to this, what is, let's just put in things that users might type in to an LLM. Which I think is a silly place to start because one of the most common things people might say is reply to this email. That's a very popular thing that a person might put into an LLM. There's no context in there. And the thing is that an LLM works on the context. So that gives us another problem, which is do you start tracking the APIs, which kind of goes into the LLMs at a core level and takes out any of the user, you know, the user sort of prompts, user attributes, or do you try and scrape and try and make it a bit more like the users was going in and accept then that you're, apart from breaking terms of service, probably, you know, you're going in and causing some challenges, but you're getting cheaper results, but they may be a bit more like the human. So I think that when it comes to what the users type in, it's a different issue. And absolutely, we do track that, but I wouldn't start there. In fact, when we launched the product, we didn't have any of that stuff in there. I think, and yes, you've got to do different models, but also there's different models within the models. So Gemini has all these different versions, et cetera, and you can just get out of hand very, very quickly. So you've got to just take a basket of what you think is going to be good and sensible and give you direction without taking so much that you're just spending money and burning up the planet for no good reason. So I think that the sensible prompts to track are ones that are generic in nature and encouraging lots of brands to come back to. You never put your brand in a user prompt. You're trying to find out what the AI, you know, you're trying to find out at that point whether the AI recommends you when the user types in something that you would hope to be recommended for. But if you put your own brand in the question, then of course you're going to be coming in the answer. So you need to be selective on your prompts that you track and you need to put them around your business, not around what's the most common phrase or search volume, because there is no common search volume, frankly. It's a misnomer. What people are actually doing is they're going through a long conversation with an LLM and then it says, what about hotels, you know, at that point, you know, and it already knows that it's in Brighton or whatever it may be. And they already know that they want luxury or budget or whatever. So we can't take the prompts as gospel and we shouldn't. We should take the prompts as indicative of ideal customer profiles that we would like to see, you know, come back with our brand in their responses. And that's a better way to do it. So prompt tracking versus what AI knows about you, two different areas of the same overall theory.

James Dooley: Yeah, for sure. For everyone who's listening to this, I think what AI knows about you is very key. I think what needs to happen is you need to define the entity and have all the entity attributes of exactly who you are, what you do and why you're brilliant. At that point, because it is probabilistic, what you want to try to do is that every single prompt that's going in, a lot of them are very unique in the way that they're doing it. It's looking at personalization, like you said there with regards to budget and premium in hotels. And then what's able to do is if you can map out everything and they understand exactly who you are and what you do, at that point, you're hoping that what AI says, what it says about who you are, but then also on the output, who it's recommending. It's recommending you because they've got that confidence and clarity to understand these are the best entity for the question that you're doing. Dickson, it's been an absolute pleasure again. And for anyone who wants to see any more episodes, make sure you check out the link in the description. There's more plus episodes where I'm going through about entity SEO and how to get sighted in the large language models. Thank you very much.

Dixon Jones: Thanks Chase.