Tom Rudnai, founder and CEO of Demand-Genius, explains why AI shapes most of a B2B buyer's shortlist from memory before any search happens. In the Dark AI study, AI retrieved and cited a brand in 0% of awareness prompts, 0% of consideration prompts and 48% of conversion prompts. He covers intent matching, what agency leaders said about selling GEO, how to measure GEO with self-reported attribution, branded search and win rates, and why citations are not a good proxy for influence.
Most GEO dashboards treat citations as the north star. Tom Rudnai, founder and CEO of Demand-Genius, thinks that measures only a sliver of what AI does to a B2B buying decision. His team's Dark AI study ran prompts at every stage of the funnel, and at the awareness and consideration stages the models did not retrieve or cite a single brand. By the time a buyer runs a search, AI has often already shaped who makes the shortlist.
Demand-Genius ran a large set of B2B prompts at three stages of the buyer journey: awareness, consideration and conversion. The question was simple: when does AI go out and retrieve content, and when does it answer from memory? At awareness, the retrieval and citation rate was 0%. At consideration, also 0%. Only at conversion did it rise to 48%. Tom's conclusion is that if citations are your main metric, you are only measuring the bottom of the funnel. In a transactional sale that may be enough. In a B2B purchase with a six to twelve month buying cycle, that final prompt is a very small part of how buyers use AI.
Tom pushes back on that refrain. Search is a tool in AI's toolbox, so search visibility will keep influencing AI answers for as long as models use it. But for large parts of the conversations buyers have with AI, the model never searches at all. It answers from its parametric knowledge, which is a completely different problem. The upside is enormous: if a piece of original research becomes part of how a model understands your category, the model repeats it to every buyer who asks. The downside is that it is slower, harder to measure and less predictable.
Models are not constantly crawling the web the way Google does. A trained model can invoke retrieval at answer time, which is why SEO can move AI answers quickly. Changing what the model itself knows means waiting for the next training run, with no guarantee your content makes it in. Tom's advice depends on your goals: if you are building a category-defining brand over years, memory is where the leverage is. If you have a lead target to hit in six months, it probably is not your first priority. Paris adds that this is why GEO has more in common with brand marketing than with performance marketing.
One signal Demand-Genius watches is the mention-to-citation ratio: how often a brand mention comes with a citation. Mentions backed by citations suggest visibility driven by retrieved content. Mentions without citations suggest visibility driven by brand weight in the model's memory. Tom is clear it is not a perfect judge, because AI does not always cite its sources, but in aggregate it gives a useful signal. He also floats a question he wants to research: as a brand becomes strongly associated with a concept, does AI start searching for that brand by name, a kind of branded search from AI?
Tom calls it intent matching. At awareness, the model is exploratory in tone, names few brands and does not retrieve. At consideration, its language turns comparative and the brand count rises, but it still does not go out to corroborate. At conversion, it becomes directive, brings in many more brands and starts checking external sources, which in B2B often means product pages, pricing pages and G2. The details vary by category, but the pattern shows how closely AI tailors both language and sources to what the user is trying to do.
For the first time, Demand-Genius surveyed humans instead of models: agency MDs and commercial leaders, mostly in the UK, across SEO, PR, content, brand and influencer agencies. Where PR agencies have a named GEO package, it is usually led by someone brought in from SEO. GEO budgets are mostly being carved out of SEO and performance marketing, which Tom sees as a risk: it makes GEO an easy sell as a new traffic channel at the moment CMOs face diminishing returns everywhere else. He also found that teams learn GEO mostly from LinkedIn and social media, with research at the bottom of the list. What travels well on an algorithm is quick wins and hacks, not nuance, and everyone ends up in the echo chamber of their own discipline.
Some AI traffic is still directly attributable, and it is worth tracking. Beyond that, Tom's stack is self-reported attribution (an open "where did you hear about us?" field on forms and calls), uplift in branded search and, for B2B, improving win rates later in the funnel. His example is the CFO who goes into Gemini or Claude before signing off a six-figure deal and asks whether there is anything to worry about. Whether AI says yes or no at that moment matters more than being mentioned to someone casually researching tools. Sales reps already report long lists of AI-generated questions from prospects. The job of GEO is to answer those inside the AI conversation, so the decision comes easily in your direction. That work needs buy-in from the CMO or CRO, because it touches positioning, content and reputation, not just SEO.
Search rewards well-optimized content: find a keyword, answer it well, rise up the rankings. AI doesn't care how well optimized an answer is, because it already knows the answer. It wants something new that is not in its training data. Tom argues content strategy needs to shift from roughly all new content toward a much more even split with cultivating your whole digital footprint, including auditing an existing library for information gain, narrative strength, accuracy and currency. Brands that built huge libraries over twenty years of good SEO work now have content telling an outdated story about who they are, and AI sees that gap in a way Google did not. Tom is skeptical of studies claiming AI prefers the top of the page; good writers have always put the most important material first.
No, says Tom. Just because a piece of content does not show up in citation data does not mean it is not shaping answers, though absence does not prove influence either. His closing advice is the jellyfish model. Viral hacks will keep coming, and if you are not careful you end up playing Whac-A-Mole, chasing whichever source is getting cited this week. The tentacles are the channels. Put your energy into the body: high information gain content, presented compellingly across as many surfaces as AI reads.
Find Tom Rudnai and Demand-Genius at demand-genius.com, or connect with Tom on LinkedIn.
Paris Childress: Hi, everybody. Welcome back to The GEO Show, brought to you by GEOforge, full self-driving for AI visibility. Today my guest is Tom Rudnai, who is the founder and CEO of Demand-Genius, which helps B2B brands and agencies understand how AI models present, present them to buyers. He came to AI search from digital media and B2B SaaS, and his team publishes original research on how models actually behave across the buyer journey, including the Dark AI study, a study of how conversational context changes AI answers, and a new survey of agency leaders on selling GEO. So Tom, welcome to the show
Tom Rudnai: Thank you for having me.
Paris Childress: Well, look, with that intro, I, I definitely wanna get right into some of your research work and kick off with, You, you did, you, you published something called a Dark AI study. can you tell us more about that?
Tom Rudnai: Yeah. So it's one of-- We, we've done kind of two particular pieces of research that I'm really, really proud of. I really feel have kind of moved forward the conversation a little bit around GEO. So the Dark AI one, we specialize, and my background is always in B2B, so I sh-- I always feel the need to caveat that a little bit with, with, with where a lot of my perspective comes from 'cause I do think there are nuances to GEO always between based on the type of company you are, the type of go-to-market, the role that you-- the AI needs to fill for you, whether it's lead generation or conversion, kind of agent-to-agent, transactions. and we, we come at it from the B2B perspective. But what we were looking at in the Dark AI study was we ran a whole load of prompts at awareness, consideration, and conversion, so the three broad stages of the marketing funnel. And we wanted to understand how does AI behave at each of those stages and how does it differ in terms of when it invokes retrieval, when it answers from memory, so to speak, how it cites brands and brings brands into the conversation. And what we noticed was really, really interesting. So we noticed when you look in each of those three stages, in awareness, the retrieval and cite-- therefore, citation rate was zero percent. In consideration, zero percent. And in, conversion, forty-eight percent. So when you tot that all up, it means percent of the prompts that we ask at different stages of the buyer journey actually invoke retrieval and cite a brand directly. So in the context of a market where we have increasingly, almost by default because no one bothered to think of anything better, so there we get controversial, made citations the kind of north star. You're actually only measuring this tiny element of the buyer journey and of the impact that AI has on your overall buyer, o-on your buyer's overall decision-making process 'cause you're limiting your, the scope of your influence if that's your, your North Star metric to those bottom-of-funnel conversion-oriented prompts back to the B2B focus, right? Because if you're in a transactional sell,
Paris Childress: Mm-hmm
Tom Rudnai: The job is being visible at the point of transaction. If you're in B2B and it's a six, twelve-month buying process, that, you know, that final prompt is a very, very tiny component of the overall job and the, the way that your buyers are using AI
Paris Childress: Yeah, that's very interesting. So all of the cita- the, the retrieval and, and of course, the citations are happening only at the bottom of funnel with high intent prompts, but, but actually it's zero, 0% at the top and 0% at the middle, right?
Tom Rudnai: Yeah. And that's always my pushback a little bit, the, the, the old refrain of GEO is just SEO. Like there, there are, there are a lot of differences and there are of course, S- search is a tool in AI's toolbox, right? So your search visibility will always have an impact on you're showing up in AI for as long as that's a tool in their toolbox.
Paris Childress: Yeah.
Tom Rudnai: But for huge parts of, of, of the, the conversations that your buyers are having with AI that you want to influence, it is not going to search at all. It is answering from memory and it's a completely different problem
Paris Childress: Yeah. So then how do, how do brands have any influence in top or middle of the funnel when there is no retrieval, there, there are no citations, there probably won't be any brands mentioned 'cause it's educational maybe? should we be even bothering trying to create content to, to train AI in, in those stages of the funnel?
Tom Rudnai: Yeah, it's a good question, and, and I-- so I don't have a final answer for you there, and that's one of,
Paris Childress: Mm-hmm
Tom Rudnai: Of building in this space. You have to get very comfortable saying that you don't know things and, and basically saying, "I'll get back to you." I, I… No, I don't think we should not bother. I th- I think there's, there's two things. One is the opportunity and the risk is too high to not bother. so I ju- I just think it g- it goes so overlooked, the potential that AI can offer, when if-- when you go further up the funnel, right? What you have in AI is the opportunity to embed a piece of research, a piece of content, a perspective that you have in the parametric knowledge of a model so that it literally says it to every buyer. It can become like part of the, defining knowledge of your category. and that's so powerful, right? Th-th-there is the opportunity for a piece of content to get literally unlimited distribution within your category because if you can make it s- part of the way that the model understands that category, then it's going to the way that it presents it to absolutely every buyer, and that's too big to ignore. Now, it's much harder to track and measure, and it's much slower to influence. So we know that models, that models are rebuilt often enough, and they give you certain windows where you can become part of the training data for a new model, right? And so it's always gonna be slower because you are waiting for that, and it is a black
Paris Childress: Mm-hmm
Tom Rudnai: An extent. But these, OpenAI, Anthropic, whoever it is, they are desperately going out and searching for data to train them on. If you create original research content, and I know we're gonna talk about information gain in a little bit, that tends to be the thing that we see have the most impact on AI visibility.
Paris Childress: Mm-hmm
Tom Rudnai: Really powerful information gain content that genuinely moves knowledge forward in your category, will latch onto that, and it can form part of its parametric knowledge, and that's what gets you some influence earlier on in the buyer journey when it's not going to retrieval. there's a metric we look at, which is not-- There's no perfect metrics in this world, but there's a metric we look at which is, like mention-to-citation ra-ratio, which is one of the things we use to understand this, which is basically how many of your mentions are followed up with a citation, and that helps us understand, okay, is your visibility being driven largely by content, therefore, i.e., gone and retrieved a piece of content, cited it, and mentioned you alongside that, or is your visibility being driven by brand influence and brand weight,
Paris Childress: Mm-hmm
Tom Rudnai: Mentioning you, but there's not content always to back it up. And we're often looking at that to give a better idea of which of those two things is driving your current visibility.
Paris Childress: Interesting. So if you see a brand mention that is not, it's not mirrored by a citation, but, but you get a brand mention, that's most likely then coming from an AI's parametric knowledge and more influenced by, long- longer term training data as opposed to, real-time retrieval. Is that… Did I get that right?
Tom Rudnai: The hypothesis.
Paris Childress: Mm-hmm.
Tom Rudnai: It, it's not a perfect, it's not a perfect judge. It's not the, the AI doesn't always give its source for everything that it cites. but, but that's the, the kind of hypothesis, and certainly if, if you look at it in the aggregate, then I think you can get some, you can get that signal from it
Paris Childress: Yeah. And does the reason why it takes long to, to influence an AI's parametric knowledge, is that just because they're, they're updating their training data, maybe every few months or two or three times a year? The, these large… Th- they're doing these big training runs, n- not, not every week or even every month, but maybe every few months
Tom Rudnai: Yeah. There's-- I think it's one of the individual models, they are trained and they're not constantly going out and kind of indexing the internet in the way that Google is. It's one of the things that
Paris Childress: Mhm
Tom Rudnai: See kind of search people talking about is like, will that content get indexed by AI? Like that, that, that, that as a sentence does not make any sense.
Paris Childress: Mm-hmm.
Tom Rudnai: It, it, it doesn't go and do that. Opus 5.5, for example, is not constantly going, scanning around the internet and looking for new things that it can add into its knowledge. Opus 5.5 is a model. It can invoke retrieval, and that's where it can discover something. So that's why your SEO can have a faster impact because that's how you influence how Opus 5.5 responds to a buyer. But your
Paris Childress: Mm-hmm
Tom Rudnai: To influence Opus 5.5 is, is kind of gone at this point. You're waiting for 5.6. now we
Paris Childress: Mm-hmm
Tom Rudnai: Come out fast, but that's the challenge. It's slower, it's undefined because you can't guarantee yourself that you're gonna be in the training data of whatever latest model is trained on, right? You're, you're
Paris Childress: Mm-hmm
Tom Rudnai: It's slow and it's not direct, it's not clearly predictable, but it is super, super high in impact. so it all comes back, and that's why there's so much nuance around GEO, right? And that's why I really… I can be quite controversial 'cause I get very fed up with seeing kind of hype and t- and quick takes on LinkedIn of, you know, do this and you're, it's your magic key to visibility. Like a lot of it comes back to what's your goal as an organization? What payback do you have to show over what time horizon? Are you looking to build a category-defining brand? In which case everything that I'm talking about now is absolutely something that you should be thinking about. Or do you have a lead target to hit in the next six months? In which case probably don't bother. like
Paris Childress: Yeah
Tom Rudnai: Many, so many ifs and buts
Paris Childress: Yeah. I, I actually think that GEO is, has more in common with brand marketing than with performance marketing, and that that typically means that you have to accept a, a lesser degree of, measurability and attribution, and that you need to extend your goals out a, a little bit longer to a longer horizon. Because if it's fundamentally the, the goal is to, position your brand in the mental model of AI, and I think that's more in line with your, your term of parametric knowledge, which is that we, we wanna cement AI's knowledge about our brand and how it associates our brand with key concepts, key entities, and, and if we focus on information gain, that might not get us ranked high in SEO and, and it might not give us the instant, retrieval and citation deliverability in the short term, but we're then positioning ourselves to contribute more to AI's mid to long-term parametric knowledge. Which means that as over time, AI is gonna lean more and more heavily on its training data and less on search retrieval, and it already knows that, let's say it, Demand-Genius is known for this thing, and when that comes up in a conversation, it'll feel confident to deliver your brand as a part of the answer, and it won't matter whether you rank a piece of content in Google for that, because there isn't going to be a search retrieval process. I, I actually do believe that over time as-- I, I think the windows of, of training data updates by these models, I think these windows are gonna get shorter and shorter. And as they get better and better at it, I think their goal is to-- I, I think that the LLMs would prefer to deliver an answer from, a confident answer from their training dataset as opposed to having to, to, to, to go through a tool like search and do a retrieval, even though they do that very efficiently. I think they would prefer to, to really have all-knowing knowledge of the world and then just to be able to deliver that answer straight away. So I think what you're saying and, and what your Dark AI study is alluding to is that the best investment for, for the long term for parametric knowledge is to deliver high information gain content that positions you to train, to, to be able to train AI so that its next, its next training run will see it, updated, and then m-make you the brand for a whole, a whole topic area perhaps.
Tom Rudnai: There-- Just quickly, there's so much interesting there. You said one th-- Like, I love the i-i-- I've thought about this a little bit, and I'd love to do some more research into it, is the concept of, like, almost branded search from AI. So if we can improve our
Paris Childress: Mm-hmm
Tom Rudnai: Weight and our, our kind of association with the concepts that we want to be in their parametric knowledge, do they actually just go start searching for Demand-Genius rather than searching in Google for options? And almost is, is
Paris Childress: Mm-hmm
Tom Rudnai: Thing that we can start to measure? I've
Paris Childress: Mm-hmm
Tom Rudnai: Nothing to share, but it's just a fascinating concept.
Paris Childress: Yeah
Tom Rudnai: On the topic of whether they ultimately-- I do think they will ultimately try and be more selective over when they invoke retrieval because we are gonna see over the next year AI, the, the frontier labs be a lot more conscious of token cost. and I think that will, will, will lead them to, to be a bit more selective there.
Paris Childress: Mm-hmm.
Tom Rudnai: Go up, generally speaking, as something that is more commonly used because it produces a better product for the user. But I think their, their focus is gonna be moving away from producing a great product for the user and towards making some money soon. we
Paris Childress: Mm-hmm.
Tom Rudnai: Ob-observed something in the Dark AI study where as you move down the funnel, there's this complete change in the way that the model responds and the way that it thinks as well. It did something that I've, I ca-- I kind of call intent matching. So it's really smart at basically understanding what the user is trying to achieve. So a, a top-of-funnel awareness query, its behavior is very, very exploratory, and the language that it uses is exploratory because that reflects the intent that sits behind the question. And you can see that in
Paris Childress: Mm-hmm
Tom Rudnai: That it uses and also in the brand count. It's not bringing in lots of brand. It doesn't invoke retrieval. Basically, it's kind of freewheeling it, a bit like I am now. It's just kind of talking,
Paris Childress: Mm-hmm
Tom Rudnai: From the hip. As you go down the funnel, it in consideration, it then becomes very comparative in its language, and the brand count increases. So it starts bringing a few more brands into it, but it still doesn't go and kind of corroborate what it's saying, and that's where we observe the zero percent retrieval rate. When you then get to those conversion-oriented queries, we see the brand rate go from one point one seven, in to six point three three. So suddenly it starts bringing lots of brands in. It becomes very directive in its language, so it helps support the user in making a decision, right? Because it recognizes that's now what they're trying to do. And it starts going and corroborating its information externally, and that's where we see a lot of G2 coming into citations in our world. So don't-- Whatever world you're in, this is gonna be a slightly different picture, but a lot of
Paris Childress: Mm-hmm
Tom Rudnai: Product pages, pricing pages, G2. and I just think that's really interesting. So it, it fits what you're saying very, very clearly, that AI is really smart at tailor its behavior and its response in every way from the lang-- down to the language it uses to the
Paris Childress: Mm-hmm
Tom Rudnai: You're trying to achieve
Paris Childress: Yeah. Yeah, very interesting. And I, I, I can say I've observed that just with my own, behavior. It, it does a good job of understanding where you are in the funnel and, and what your intent is. Yeah. Let's move on to the other, the other study that, we touched on in the intro, which is a survey of agency leaders on selling GEO. La- tell me more about that
Tom Rudnai: Yeah. I was, I was proud of this actually. So all of the, all of the research studies that we've done to this point have been talking to AI, rather than talking to humans, which it turns out is way easier, and way quicker. but we, we, we took this on where we wanted to interview a whole load of agencies. I think generally speaking, the sense that I get from speaking to people in this space is that we're in a very hype-driven market and no one knows who to trust. And there, there is… The, the overwhelming I get myself is just one of complete uncertainty, confusion. And so we were trying to speak to agencies about how they're going about pricing, packaging, delivering GEO in amongst all of that uncertainty. so
Paris Childress: Mm-hmm
Tom Rudnai: To about 50 different agency MDs, mostly here in the UK, MDs and commercial leaders. a pretty extensive survey where we got a whole load of awesome data in terms of how they package it up, who delivers it within the agency, how they approach upskilling their team, client responses to it, and also some great kind of qualitative feedback and did some, some recorded interviews.
Paris Childress: Yeah.
Tom Rudnai: That was, that was really great.
Paris Childress: Let, let me take a stab at what, were some of the predominant findings. I'm guessing that most, GEO practitioners are, are ex-SEO people that are transitioned over. Is that what you found?
Tom Rudnai: Yeah. obviously, so we, we, we deliberately cultivated a survey across SEO, PR agencies, content agencies, brand agencies, a couple of influencer agencies as well.
Paris Childress: Mm-hmm.
Tom Rudnai: But yeah, we did ge- generally find that, and we generally found that where PR agencies are doing GEO and do have a named package, it's because they've actually brought in an SEO person to lead that. so they're very much SEO as a discipline has been the early adopter on moving on GEO,
Paris Childress: Mm-hmm.
Tom Rudnai: Which is, is interesting to see. So yeah, spot on. You're, you're one for one. What, what, what are your other predictions?
Paris Childress: F- that was, that was the main prediction, but, and I also think about budgets and also where does the GEO budget come from because if, if y- you're a CMO or, I mean, if, if you're going to the CFO to say, "We need extra budget," or, "We need a new budget for this new category," does it, does it get carved out of SEO? Does it get added on top? did your survey address that at all? H- how does it get funded, basically?
Tom Rudnai: Yeah, it did. And that's actually one of the-- I think that's one of the big challenges I see for the industry going forward off the back of it actually. generally speaking, it's being carved out of SEO and performance marketing budgets, and I think that there's a lot of reasons for that. That's what the, I'll say, the, the snake oil salesmen that exist in this space want to do that because it makes it an easy sell, right? You, you, you say you're gonna get traffic and leads from this new channel, and it, it's very easy to, to, to, to sell. so I, I, I think there's also… And so that's driving it from the kind of, GEO side, the supply side. From the demand side, that's exactly what every marketer is desperate for. Like, what are we seeing in the mark- if, if in the market at the moment if you're a CMO? You're seeing basically diminishing returns from paid, diminishing returns from organic or like SEO, paid, paid is becoming incredibly crowded. Even go to things like direct outreach and outbound. all of these things are giving you diminishing returns, right? Every channel that you work. So you are desperate for this magic bullet new channel to come in and be your kind of golden ticket to reaching
Paris Childress: It's a-- Yeah, it's like a lifeline, yeah, for people, yeah.
Tom Rudnai: It--
Paris Childress: As everything else is is decaying and going down, this is the new shiny object
Tom Rudnai: So it's like the perfect coalition of desperation and lack of moral fiber that is allowing people to the kind of more snake oily people within the industry to, to do that. and I, I should say that there's loads of people trying to do fantastic work in it, so to, to, to, but it, I, I don't mean to be too kind of negative or… but so yeah, I think that's one of the great conflicts or contradictions that we see is as a result f- and because GEO just sounds so damn close to SEO, we're seeing
Paris Childress: I was gonna say, if we're calling it-- we're giving it the same naming convention, which might be a mistake. It's a, it's another thing that's getting optimized, but I, I'm not sure that what we're doing is optimizing for an algorithm anymore. You know, it's, it's quite different. But yeah, the, the acronyms are starting to stick, and they probably will stick. And it's, it shares two, two of the, two of the three letters of the, of the old acronym. So people will definitely associate it with, with SEO
Tom Rudnai: And I find it so interesting how it's-- you can see the impact that the general information ecosystem that we're building this new discipline in has on the way that it's taking shape as well. So one of the most interesting findings I thought from the survey was we asked where their team, them and their team are learning about GEO, and I, I, I forget the exact numbers, but around about thirty percent, the bottom one was actual like research, and the top one in the high sixties percent was LinkedIn, social media,
Paris Childress: Yeah
Tom Rudnai: So everyone is learning about GEO in an echo chamber, and they're learning about it from algorithms. Now, what travels well on an algorithm? Is it a deep, nuanced view that requires you to rethink how you think about outputs and delivering work? No. It's quick wins. It's hacks. It's here's the
Paris Childress: Mm-hmm
Tom Rudnai: Channels by AI, ignoring the nuance of the fact that that's different in every category, and you're making decisions off data that has no application to you.
Paris Childress: Right.
Tom Rudnai: So that's one
Paris Childress: And by the way, you said, said it perfectly when w- it's, it's, it's an echo chamber because the content that reinforces your, your beliefs about GEO will be the content that you're gonna like and comment on most likely, which means the algorithm is gonna keep giving you more and more of that. and then it becomes a reinforcement loop, unfortunately. I mean, similar to, to Facebook and, I mean, you could make a political, analogy also on Facebook that people, people tend to like and engage with, content that reinforces their worldview, and then the algorithm is gonna keep serving you more of that, and you just get more and more entrenched in your, in your set of beliefs that way. Yeah.
Tom Rudnai: I see that really prominently. Like, 'cause it-- the truth about geo is it's a discipline that sits across SEO, yes, but also brand and PR and all of these different disciplines. And I kind of see it because I have a bit of a view across it on social media. You look in, like if you're in the PR echo chamber, you're seeing stats like eighty-six percent of citations come from offsite, and you're quoting these things, and you're saying, "There we go. See? It's a PR job."
Paris Childress: Yeah. It's a good argument too
Tom Rudnai: Load of other, of other data, and they, they all live in this echo chamber, and they rightly identify their role in solving the problem, but they wrongly fail to identify everyone else's, and therefore, that they are one part of a broader picture. That,
Paris Childress: Mm-hmm
Tom Rudnai: High-level view on what's going on here.
Paris Childress: Yeah. Yeah. It, it really is. It's kind of eating, eating all the other adjacent d- disciplines and absorbing them in a way. Yeah. So let's, let's talk about Brighton. This is where we were both speaking at brightonSEO a couple of weeks ago. So the conference, one of the things that struck me about the conference was that, you know, practically every exhibitor was, now-- is now pitching some sort of a AI rank tracking or a GEO tool, and almost nobody's talking about the old stuff like link building or even just content marketing.
Tom Rudnai: Yeah. And I think, I think the market will evolve to the point where now it's like ev-- th- it's such a hype cycle, and there's so much demand that if you're in and around it, you're crazy not to have an offering. But it means that
Paris Childress: Mm-hmm
Tom Rudnai: I, I think there will become a lot more focus on quality of data research that underpins it and the basically the differentiation around your GEO offer, right? There's seven
Paris Childress: Mm-hmm
Tom Rudnai: Just ticked over the other day into seven hundred GEO vendors or AEO vendors on, G2,
Paris Childress: Oh, is it above 700? Yeah. It was at 650, like, two weeks ago. It's so
Tom Rudnai: It's--
Paris Childress: Crazy
Tom Rudnai: Conversations with our investors back in about April, and I was saying like, "God, it's so hard. We've got two hundred and fifty competitors." And then you look now, and it's more than doubled since then.
Paris Childress: Yeah
Tom Rudnai: So it's, it's, it's crazy. But I tell you what I did think at Brighton, I think the conversation has moved on in a really positive way in SEO. Like I, to be really honest, got quite frustrated at brightonSEO last year and the year before because I felt there was this attitude of like, AI is this mean, nasty thing that's coming to disrupt us SEO people who just want to kind of keep doing SEO. And that frustrated me. I was like, look, it's-- we can hand-wring over it, but it, it, it's happened. We have to think about how to kind of go forward. But I
Paris Childress: Mm-hmm
Tom Rudnai: There was such a different energy this year where it starts to feel like everyone is actually excited about it as an opportunity for here's something that we can get good at and that can be a… something we do going forward. And I thought that, there
Paris Childress: Yeah.
Tom Rudnai: Nice tone shift, I thought.
Paris Childress: Yeah, I wasn't at the, any previous Brightons, but, the, the, the tone I thought was, was great, and people are, I think, looking at this now much more opportunistically.
Tom Rudnai: Mm-hmm
Paris Childress: Yeah. let's get back into some of the research and, and data side of things 'cause I have a c-couple of other, things that I wanna touch on. in terms of the measurement, this is, this is one of the more difficult aspects of GEO now. How do we measure ROI? 'Cause we, we are losing the click, which was the, the critical bridge that showed how investment and efforts, translated into web traffic and conversions, and usually those conversions had some monetary value that allowed us to report on return on investment. Now we can't really do that, at least not right now. How do you think, we, we as a, the stewards of this new category, how do we deal with this? Because if people are equating this, non-marketing people that are, that hold the purse strings, like the s- the CFOs, CEOs, they're, they're going to equate this as the, the, the new SEO. I mean, it's, it's inevitable. And then the old SEO gave me a R-ROI approx- approximate ROI calculation, and now the new version of this called GEO cannot. How do we, how do we overcome this, and how do we c- how do we get investment and budgets internally?
Tom Rudnai: Yeah. And I think it, I think it's a balance between making sure that you're trying to deliver against what your client or what your boss actually cares about, and trying to also manage expectations as to what the returns and what, what the impact is going to be. I think, look, so you-- there, there is still some within the existing funnel that we can track and measure, right? So you will
Paris Childress: Mm-hmm
Tom Rudnai: Get directly attributable traffic-- attributable, I think that was right, traffic that comes from AI and that you can follow through, and that's fantastic. The other things that we see people do, and it's normally the best way of attributing performance back to it, is self-reported attribution. So
Paris Childress: Mm-hmm
Tom Rudnai: Asking customers either on the, the call or in your kind of sign-up form, "Where did you hear about us?" Make sure that that's a kind of open field generally, and, and you get really good insight into what their… It's not perfect in terms of genuinely what was their first touch. Where did they hear about you? But it's actually what was the most important one, and that, that tends to be very good for AI 'cause it tends to be very high impact and high intent. so that was what we would like set up with a client. The other things that you can look for is you also think about, okay, how do we measure brand? and that's things like branded search and stuff like that. So and you would expect to see an uplift on that.
Paris Childress: Mm-hmm
Tom Rudnai: I would say to a client, "If we're doing a really, really good job of GEO, yes, you should be seeing some traffic and, and leads gen impact from it. You should also be seeing general uplift in things like branded search, and actually across the whole funnel," and come back to a B2B world, "but you should be im-- you, you should be seeing improving win rates later down the funnel." Because what's
Paris Childress: Mm-hmm
Tom Rudnai: A B2B buyer is about to buy a product? Their CFO is going into Gemini or Claude or whatever, and it's saying, "I'm about to buy Demand-Genius, anything I should worry about?" And that's the most high impact thing in any of this, right? Not whether you're mentioned to the, know, the, the, the,
Paris Childress: Ja
Tom Rudnai: The, the accountant that's looking for tools, but whether when the CFO goes and validates it before signing off a six-figure deal, does it say yes or no?
Paris Childress: Yeah. It's a sales assist--
Tom Rudnai: Uplift everywhere
Paris Childress: It's kind of, it's kind of a sales as-assist channel that if, you know, it's kind of the, the pre-sa- maybe a pre-sales rep before somebody's gonna go and book a demo. They're, they're gonna ask a lot of those questions to AI. They're gonna get some answers. They might even validate, try to validate those same questions with the actual sales rep on the call. But in a lot of ways that if, if GEO has been done well, then the LLM acts as a, a fantastic pre-sales person, and can, yeah, they can definitely improve the, the win rates. Yeah
Tom Rudnai: Yeah. A-and we're seeing more and more of that, of marketing, like just e- even just in kind of communities that I'm in, marketers talking about how their sales reps are now saying that they get this long laundry list of questions that have obviously been generated by AI, and they have to answer. And so that's great stuff that if you're working on this, you need to get that information because your job basically is to stop those questions getting to the sales rep, but make sure that when they're having the conversation in AI, they just get answered and, and, and it makes it really easy for people to, to make a decision in your direction.
Paris Childress: Ja.
Tom Rudnai: I th-
Paris Childress: Ja.
Tom Rudnai: One, one thing we've learnt through this process is we have to work at a senior level, right? So we have to have real engagement and buy-in from the CMO or the CRO, and I think that's something which if you're doing this in-house, you can probably take on board as well. It, it's very hard to do a good job of GEO in a silo as an SEO person or with an SEO remit. It kind of needs to go above, and you, and you need buy-in to make genuine changes across positioning, content, and reputation.
Paris Childress: Yep. Yeah, absolutely. Man, great, great stuff. Tell, tell me more about Demand-Genius and the platform. How did you all, develop it? You mentioned that you got in a little bit early when there was only a couple of hundred competitors, or may- maybe it was even much earlier than that. I think that was just when you had last, mentioned it to the investor. But t- tell me, tell me more about Demand-Genius. How does the platform work? What, what makes it different and unique?
Tom Rudnai: Yeah. Well, it's funny. but we, we certainly weren't one of the absolute first on it. w-we originally were working much more in the content marketing space. and then,
Paris Childress: Mm-hmm.
Tom Rudnai: Always think of there's, there's someone told me, a couple-- a little while ago that about eighteen months ago, any MarTech brand that didn't have product market fit became a GEO brand. And really, to be honest, we're one of them.
Paris Childress: Yeah
Tom Rudnai: So we were working in content, and we, we, we have this platform which can help measure the revenue impact of your content by kind of being a, hooking up web traffic to, conversion through the CRM and uses AI to do really deep, what I call it, qualitative content audits. So
Paris Childress: Mm-hmm
Tom Rudnai: Your entire content library, which is far too big to fit in an AI's context window, turns it into structured data using AI to fill out that database in a way which allows a single AI session to basically traverse the whole of your content library. so what that means is where at the moment, if you've got a ten thousand page library, which some of the biggest brands do, right now, if you just get AI to do a strategic task for you, it can look at naught point five percent of your content as an input into its strategy, and so it's doing everything completely blind to what you've done historically.
Paris Childress: Mm-hmm.
Tom Rudnai: So many quick wins and opportunities to upgrade content, remove content that's no longer doing anything or is doing damage. what we do, we turn all of that into data so that every AI session has a full context on what you've done, and it can be up to forty-two times as much of your existing content that you fit into the context when you're making these decisions. So it
Paris Childress: Mm-hmm
Tom Rudnai: Both the floor and the ceiling of every decision that you make from a content perspective with AI.
Paris Childress: Interesting
Tom Rudnai: So that's our kind of background. And then over the last twelve months, off the back of our research, we've built out a whole kind of AI visibility tracking as well.
Paris Childress: Yeah. Is what you described,
Tom Rudnai: To… No, I think
Paris Childress: Is that a, is that a vector database? I'm just curious when you, you, you talked about to be able to actually have a, an enormous amount of context, content that can be brought into a single, like a, like a knowledge base of sorts. Is it a, is it a vector database? 'Cause, that, that is a concept I understand, but is it, is it that or is it something different?
Tom Rudnai: It's something different. It stops short of that. and I know, Ryan Law from Ahrefs did a great talk at brightonSEO on this. vector database is brilliant if you're trying to actually recreate content 'cause you can store kind of bits of text that you've written and understand the relationship between all of it, and it's great for recreating content that really feels like you. because what we're doing is more about the content strategy decisions, we just turn it into a structured database, basically a, a giant database as you would understand it, right? A giant table.
Paris Childress: Mhm
Tom Rudnai: Means is that when making a strategic decision, AI can kind of traverse that and say, "Okay, we're talking about the blog here and looking at this particular gap around this audience." So we traverse it by audience and blog and kind of find the pieces of content that are relevant and then go and retrieve those.
Paris Childress: Mm-hmm.
Tom Rudnai: So it, it's kind of a mi- a middle ground for, for, for strateg- for strategy tasks rather than writing tasks is, is super powerful
Paris Childress: Gotcha. And I, I guess a lot of those strategies are pointing the marketer in, into the direction of what content needs to be updated. What, what are the priorities? So the, the certain content maybe should be removed entirely, other pieces of content sh- should be updated and refreshed. There's another group that probably could be just left alone entirely. And then there's the, the bucket of what new content should be created. Am I thinking about that the right way, or is there, are there other decisions to make when it comes to your content?
Tom Rudnai: I think you are. I think one of the big changes in the new world of AI that people need to adjust to is the old world, because you're optimizing for an algorithm, it really just reads your new content, and it's kind of this constant battle to churn out new stuff to feed hungry algorithms.
Paris Childress: Mm-hmm
Tom Rudnai: AI isn't really doing that. It sees, it, it's building a perspe-- a perception of you based on literally everything on the internet, on your site and off your site. And so the focus of a content strategy, I think, where it's always been like ninety-five to five percent focused on content, that needs to now tilt towards much more of a fifty-fifty split. You're not just
Paris Childress: Mm-hmm
Tom Rudnai: Out content, you're cultivating an overall digital footprint of which your content is the most controllable element,
Paris Childress: Mm-hmm
Tom Rudnai: That is clear, consistent, compelling, rich in information gain, and like eminently citable. And in order to do that, the way that we do a content audit needs to evolve, so it's no longer about the kind of technical checkboxes that you need to tick. It's, okay, across all of our content, how are we coming out in terms of information gain, in terms of narrative strength, in terms of authority, citability, in terms of accuracy and currency? Are there outdated pieces of content that are gonna harm the perception of us or that present our brand and our products in an old way?
Paris Childress: Mm-hmm
Tom Rudnai: Most brands are pivoting six months now, every three months now is how quickly markets move,
Paris Childress: Yeah
Tom Rudnai: All of your old content is telling a completely different story of who you are. So that's where this kind of bird's-eye view your existing content becomes really important to cultivate
Paris Childress: Mm-hmm.
Tom Rudnai: I-in its entirety
Paris Childress: Yeah, that's interesting. And I guess that ratio would change depending on how, how old your brand is. So a, a, a larger and older brand may have even more of 50% efforts needing to go towards existing content, updating, refreshing, or, or removing, versus a newer brand of like a startup, for example, they still might lean towards predominantly new content just because they don't have… They just don't have a library. Mm-hmm.
Tom Rudnai: Yeah, absolutely. I mean, our, our record is I think we worked with one site which is about eleven thousand pieces of content. and you can imagine for a lot of brands, they've built that library of eleven thousand pieces of content over twenty years of doing really good search work, right?
Paris Childress: Mm-hmm.
Tom Rudnai: B-built content that reflected their positioning at the time. They repositioned, they did it again. But what happened is over time, a bigger and bigger gap opens up between who you are now and who you said you were twenty years ago. And AI kind of sees that gap in a way that Google doesn't and, and so they've never done that job of going up and clearing up behind them.
Paris Childress: Yeah
Tom Rudnai: What, what you need from content is different now, I think at a foundational level, right? Search rewards really well-optimized content. So content that finds a high intent or high volume keyword, answers it well, and you rise up the rankings. That's search in a nutshell. AI doesn't care how well optimized an answer is. It knows that anyway. What it wants is something that says something new, something differentiated that it doesn't have in its training data. That's where we talk about information gain.
Paris Childress: Yeah
Tom Rudnai: Your goal now is to differentiate your content library, and that's a totally different thing when you've often got twenty years' worth of content that is really well optimized. It's good content, but it's woefully undifferentiated. and that's what I think the big-biggest challenge for brands that have built a big SEO machine is because they need to make that shift from churning out good answers to questions
Paris Childress: Yeah, keyword optimized content
Tom Rudnai: Yeah, to building interesting, meaty, thoughtful pieces
Paris Childress: Yeah. I think one of the biggest, one of the biggest elements that was always present in, in SEO and content marketing was keyword research was at the heart of everything. So before you would do a content brief, you would do your keyword research and understand where are the opportunities, and then you would, you would design your brief around, well, this is the way we used to do it. These are my primary trophy keywords, so my H1 is gonna include my target primary keyword, and my H2 tag and my H2.3 tags are gonna progressively go down into the, I don't know, secondary, secondary keyword targets that typically would have lower volume, but they're still, semantically connected in this kind of keyword cloud concept. But that your, your content brief would essentially, try to take a keyword cloud or a keyword cluster and organize it into a well-structured, skeleton, with, with H1s, H2s, H3s. Keywords would be optimized everywhere and, yeah, LLMs don't, don't care at all about that. I mean, I don't think that they even have a notion of, of keyword research. They're gonna do their own query fan-out process, which is different. They're gonna take a prompt and break it into, a whole bunch of different, key- keyword queries just so that they have a mechanism to do that retrieval. but they're not using any kind of a keyword tool to figure out those query fan-out terms. but yeah, that's a whole, that's a whole nother subject. But the, the whole approach to content has changed, so if you've got a big legacy library of doing things the old, the old way, which is driven mostly by keyword research, that would imply that, that there's a lot of work to do to clean that up. I mean, translating those into, just up- updating them based on where your, your company and your brand is, but also changing the formats in a lot of ways. a lot of these old blog posts naturally written, written for human readers had more of a narrative style where you, you might lead off with an introduction and you might try to, you know, to be interesting and tell, tell a story as opposed to now it seems more like just get right to the point and give AI the, the, the information it needs and the facts it needs as early as possible. I've seen some studies that's, that show that a lot of the citations are pulling from at the top of the page, like the first couple of hundred words. There's more of a weighting towards, the stuff at the top.
Tom Rudnai: I've seen that, although I'm, I'm, I'm actually deeply, deeply suspicious of those studies because I think they, I think they reflect something that people do very wrong, which is present stuff that is not proper research as research, right? It, it has long been a clear writing principle, which is called pyramid writing. You start with the most interesting stuff, you start with the conclusion, and you stack
Paris Childress: Mm-hmm
Tom Rudnai: Add the detail as you go. So yeah, of course, AI mostly cites the top because that's where you put all the interesting stuff. Like I've, I've
Paris Childress: Yeah
Tom Rudnai: Anything proven that that's because that's where AI wants to read. It's just that's where the best stuff is most
Paris Childress: Mm-hmm
Tom Rudnai: Written.
Paris Childress: Yeah, good point. Yeah
Tom Rudnai: To disprove it, but I saw that… I, I got really frustrated, actually, as I saw that take off. I was like, "That, that's what…"
Paris Childress: Yeah. That's just good writing, huh? Yeah.
Tom Rudnai: Yeah, it reflects the way we write rather than the way either AI behaves, possibly. but that was my kind of, immediate reaction to it
Paris Childress: Yeah. Yeah. It's fair, it's a fair point. And, and increasingly as the internet just got flooded and attention spans started to shrink more and more, I think you started to see writers that were front-loading the content even more, getting right to the conclusion, and more, more TLDR sections started appearing even before, AI. But now it's considered an AI best practice to, to lead off with a TLDR section. But, I think that does reflect what, what readers, readers' preferences that were developing even prior to AI's arrival. You could also maybe, make the, the counterargument, Tom, that the-- does AI already naturally discount, an article that you wrote 20 years ago that still happens to be on your, on your blog? D-doesn't it know that this is clearly something that's outdated and I should pay much more attention to the, the post that was published last week?
Tom Rudnai: See, so yes, to a degree, it seems to-- it, it, it-- a lot of this comes back to AI is much smarter than search was. and so a lot of, like, write for humans, and it's doing a better and better job at reflecting them. Yeah, so to answer your question, though, yes, it will weight things more prominently if it is recent. It seems to depend partly on the topic and how clearly important recency is to it.
Paris Childress: Mm-hmm.
Tom Rudnai: I don't have really clear data on it. but yes, there, there is an element of that, but it doesn't-- but not as strong as search, and it does still see the old stuff, and it can dent the kind of perception of the overall brand and library as a whole.
Paris Childress: Mm-hmm
Tom Rudnai: More kind of functional or, or kind of practical benefit to thinking about your content in its entirety is generally speaking, that's where the quickest wins are. it's, it's, it's quicker to refresh and improve an existing piece of content and take something meaning-- something and update it and make it now really relevant than it is to write something from scratch. So there's a, a little bit of the risk of AI taking misinterpretations from it and, and, and adding confusion and also just the opportunity of, quicker wins
Paris Childress: Yeah. Yeah, it's a fair point. I guess there's a… I, I've, I've seen some research indicating that th-there is a recency factor across the board. There, there are-- citations tend to be-- There's a freshness factor that the majority of citations are… I don't know if it's less than 12 months or even nine months or something like that, but I remember seeing some study that showed majority, majority of citations appearing in LLM responses are, quite recent actually. so I do, I do think that there is part of a, a natural bias of AI towards recency, but then if it sees something, it's, it's not, If, if you have something contradictory to something a little older that maybe contradicts, something that you've published recently, I think you still run the risk of confusing AI with that, and it might wonder is that, is the older thing relevant at all still or, And then if there's any confusion or, confusion or hesitation, that's enough of a reason to, to just go back and update that old, that older piece and just, just to align everything
Tom Rudnai: I think that's where, yeah, it's important to remember that citations is not a good proxy for influence. So just because something doesn't show up in the citation data doesn't necessarily mean that it's not influencing it. Now, that doesn't prove it in the other direction either.
Paris Childress: Mm-hmm.
Tom Rudnai: But yeah, I completely, I completely agree with that, yeah
Paris Childress: Yeah. I think the emergence of these personal AI agents like the Meta Muse, Grok Bots, and now Claude Duet has just launched ChatGPT Dots. But what changes with the human behavior when you're re-responding back to an AI assistant that is prompting you for the most part? it'll be, it'll be interesting to see. Any th- any opinion about that?
Tom Rudnai: I, I couldn't sum it up better myself. That, that, that's what we say to clients, right? Is all-- there's gonna be all of these hacks, they're gonna be very viral on LinkedIn, and if you're not careful, you'll end up playing Whac-A-Mole chasing, "Okay, Reddit's getting cited. Wikipedia's getting cited," and chasing those tactics. What won't go out of style, I think, is, yeah, high information gain content presented in compelling ways across as many of the surfaces as AI reads as possible. And
Paris Childress: Mm-hmm
Tom Rudnai: In that term and put more of your energy into that, I think of it as like a jellyfish. And you've got all the little channels, the little dangly tentacles, but put your, put your energy into the body, into the substance of what you're saying and how compelling it is, then that, that's the kind of future-proof way to do this. I find it very interesting what you said about the, the… from not really even a GEO standpoint, the fact that it's gonna transition from us prompting AI to AI prompting us. I hadn't thought of it in those terms, but that is a pretty fundamental change that's gonna mull around in my head for the rest of the day now.
Paris Childress: But interesting to think about. Tom, this has been fantastic. how about why don't you tell our audience what's the best place to find you online and connect with you?
Tom Rudnai: Yeah, I'm not the most kind of visible person online always. the main place is, well, go to our website demand-genius.com, or you can connect with me on LinkedIn. There aren't many Rudnai out there, so if you search for Tom Rudnai, R-U-D-N-A-I, you will definitely find us. yeah, that's, that's about it for me on the social media, I'm afraid.
Paris Childress: Okay. That's, that's enough. Yeah, usually LinkedIn is, is quite enough. Thank- thanks a lot, Tom. I've really enjoyed the, the conversation. You're doing some groundbreaking research. Keep up the great work. You are, are definitely doing your part to move this conversation forward, and I appreciate you sharing all that with, our audience today
Tom Rudnai: Thank you, Paris. Thanks for having me on
Paris Childress: All right. Bye-bye