Why Public Relations May Be the Secret Weapon in AI Search

Every PR professional has sat in a meeting where leaders ask you to “show up in AI”

Here’s the problem. The AEO/GEO industry has built an entire measurement apparatus around citations… the moment an AI visibly names your brand. But the research shows that moment is essentially an echo of a decision that was already made, upstream, in conversations that left no trace. 

Brands have been optimizing for the receipt, when the purchase decision happened days ago.

The deeper irony is that the work which does move the needle in those invisible conversations looks exactly like what PR teams have always done. Original research. Expert voice. Earned credibility. Third-party validation. The toolkit that has historically been the hardest to attribute is, it turns out, the one doing the most work in the AI era. The people who were always right about what good looks like just never had the data to prove it… until now.

Listen For

3:35 Why Does AI Influence Start Before the Conversion Stage?

7:41 Is AEO Really Just Earned Media Strategy in Disguise?

10:09 What Are the Three Levers That Shape AI’s View of a Brand?

13:13 Are “Hey AI” Pages and Schema Tricks Ethical or Effective?

15:49 How Can PR Teams Prove Earned Media Builds AI Influence?

Guest: Tom Rudnai, Founder and CEO Demand-Genius

Website | X | LinkedIn 

Doug

Substack | Website | LinkedIn

Are you a brand with a podcast that needs support? Book a meeting with Doug Downs to talk about it.

Apply to be a guest on the podcast

Connect with us

LinkedIn | X | Instagram | You Tube | Facebook | Threads | Bluesky | Pinterest

Transcript

Lady Emily (00:00):
In his book, The Tipping Point, Malcolm Gladwell wrote about a certain kind of person, someone who shapes how an entire community understands a problem long before anyone makes a decision. He calls them Mavens.

Doug Downs (00:17):
There's a hardware store in a small town. In their 40 years, the owner, Gerald, knows everything, not just prices, I mean everything, which paint holds up in humidity, which brand of screw strips after two years, whether the new electric lawnmower is actually worth it or whether you should wait just one more season. He doesn't advertise, he doesn't have a loyalty programme, he doesn't ask for reviews. But when someone moves to town and needs to fix something, a neighbour says, “Go talk to Gerald.” And they do. And Gerald asks them three questions and points them to the right shelf and they leave knowing exactly what they need and why they need it. Gerald never shows up in anyone's analytics, no click, no referral code, no attribution. The transaction happens somewhere else later after the decision has already been made in Gerald's store, in that conversation. Malcolm Gladwell called people like Gerald Mavens, the ones who shape how we understand a problem before we've even started solving it.

(01:21):
 They don't close deals, they set the terms under which deals get closed. Now for decades, marketers measured everything that happened after Gerald, everything visible, everything traceable. Nobody measured Gerald. New research suggests that in the age of AI, Gerald is now a language model and the brands that taught it how to think about a problem are the ones that win. Today on Stories and Strategies, we find out who's been talking to Gerald.

(02:06):
 My name is Doug Downs. Our guest this week is Tom Rudnai joining us from Bermondsey, which is near Tower Bridge in London. Hey, Tom.

Tom Rudnai (02:14):
Hey, Doug. Thank you for having me on.

Doug Downs (02:16):
And for me, Tower Bridge is the one I've seen. London Bridge is the song I sang in school. We associate them as the same, but they're different things. Tower Bridge is the beautiful bridge.

Tom Rudnai (02:27):
Tower Bridge is the very pretty one, the very old one that you can't actually walk across because you're going to have a thousand tourists taking photos in front of you. But I'm sure for you, you were just one of those thousands. So it's a lovely bridge for you.

Doug Downs (02:39):
I'll do it when I'm in London in a couple of weeks here. You're the founder and CEO of DemandGenius, a London based AI search intelligence platform. And earlier this year, you published original research that tracked hundreds of B2B buying journeys inside AI systems to understand exactly when and why brands show up in AI and more importantly, when and why they don't. So your study found that brand citations and referrals occur almost exclusively at the conversion stage. So that's the very bottom of the funnel. That's where people are taking their action. But by that point, realistically, I mean, the decision has been made earlier through invisible conversations. So walk us through what you actually observed in the data and what that tells us about where the influence is actually happening.

Tom Rudnai (03:35):
Yeah, of course. So I'll give you a bit of context first. So what we did as the study was we ran an awful lot of prompts and we split them across awareness stage, consideration stage and conversion stage because we wanted to get an understanding of how AI responses varied at each of those stages. We noticed a load of different things in the data, so there's some really interesting trends in terms of how the language evolves. So at the top of the funnel, AI does this thing that we call intent matching where it's really, really smart and it reflects back what the user wants from it. So at the top of the funnel, if you ask it a very open question, it's exploratory. As it goes down, it becomes very comparative and then eventually decision oriented. It's very tempting always to think when it gives quite a firm endorsement of your brand, that it's because it really likes you. It doesn't. It's because the question actually asked it to be decision oriented and asked it to be firm. And so it helps you make a decision. So there's some really interesting trends in that that can give you some clues as to how to optimise it. What I found most interesting though is when you look at, there's a few things that you can track in that mention rate, retrieval rate, citation rate. So mention rate is how often a brand gets obviously mentioned. Citation is how often a brand gets specifically cited and linked and retrieval rate is how many times the AI actually goes looking for an answer because obviously sometimes it answers from its own training data and sometimes it goes and searches to enhance that.

(05:06):
 And what we noticed was as you go through, citation rate and retrieval are the particularly interesting ones for the purpose of this conversation and for the purpose of AI strategy. Awareness stage, both of them 0%. At consideration stage, both of them 0%. So the AI is answering from memory, it's exploring, it is not going looking for answers. So it's always very tempting to think of this as a search job. Well, it can't be if the AI doesn't search.

(05:24):
 It undermines quite a lot of how we tend to approach the problem. And then even at conversion stage, only 48% of the time, and that equates to 16% of all of the prompts that we ran, does the AI ever go looking for an answer and ever actually cite a brand? So if it's not looking for answers at these points, then where is influence being shaped? Is it being shaped much, much earlier? Because what we notice as well is if you look at the way that it surfaces brands within these responses, again, at the top of the funnel, it's very, very explorative. And it surfaces a broad array of brands. And then again, as it goes down, it narrows down. It converges on the brands that it's going to ultimately recommend against that bottom of funnel query. So what's more influential? Is it optimising the bottom of funnel prompts where the answer is largely already decided or is it actually optimising the direction in which it converges so that feeds towards your product, what you're good at, what you're bad at?

(06:35):
 Optimising for how it frames the problem, how it builds requirements for people. And that's what we found and that's what was really interesting.

Doug Downs (06:44):
Indeed, indeed. And I find myself, for lack of a better term, which there must be, trying to get inside the LLM's brain, right? Because we work in PR and essentially I've spent copious years trying to get inside people's brains. Now I'm trying to get inside the AI's brain. Information gain. Ultimately, an LLM is trying to provide that because that satisfies the user and you describe information gain as content that genuinely contributes new knowledge rather than summarising what already exists. That's a key point. And you say what shapes how LLMs understand a problem space. So how much of what you're describing as the solution is actually what I call earned media strategy or original research, expert commentary, credentialed third party coverage. Where is the influence coming from?

Tom Rudnai (07:41):
Yeah, it's funny hearing you talk about getting inside the AI's brain because I feel like that's what I spend literally every day of my life doing and it makes it very difficult to interact with humans. But I think it's a really interesting framing actually, and I think you could quite accurately characterise all of AEO or AI search optimisation as it's about earned media. You just have to look at the LLMs as an active participant in that. They are one of the earned channels. And I think if you frame it in that way, that's going to be much more helpful than a kind of SEO style framing.

(08:12):
 You explain it very well. I think at its core, SEO and the content marketing game has always been about finding a very high intent or high volume query, summarising knowledge against that query and the best summary wins, right? That's what search was in a nutshell. I can go into AI and I can get it to summarise knowledge for me on any topic within the corpus of human knowledge in a moment and it's going to do that bespoke to me, understanding a lot of my context. It's better than any content that anyone produces if that's all they're trying to do. So what we encourage a lot of our clients and a lot of people to do is focus on what makes you citable. If there's no reason to cite you in the content that you're putting out, then why would you ever get surfaced in AI? Why would you ever be able to influence the way that it thinks about not just your brand but your category as a whole?

(09:02):
 And that's where, okay, if you can just produce meaningful information gain, meaningful information gain that moves the collective knowledge of your category on a particular problem space forward, then there's something citable. There's some reason for the AI to not just surface you when it's asked directly about you, but in all of those problem framing conversations that we know are so important. And that's where we've built this information gain framework, which really helps marketers as they build that out, where you actually try and score your content and say, “Okay, are we consistently producing this high value stuff?”

Doug Downs (09:33):
So I want to be original. I want the AI to ultimately say, “Well, according to Doug Downs from Stories and Strategies, blah, blah, blah.” And that could be original research that I've done. It could be my expert commentary, see my LinkedIn channel that I've provided or what someone else has said about me. If Tom writes and says, “Well, Doug Downs at Stories and Strategies says that too is powerful,” which is the most powerful? All of the above or do I need tick boxes everywhere or should I focus on one as opposed to the other?

Tom Rudnai (10:09):
I think what seems to at the moment provide the most impact is reputation and that obviously is probably music to everyone who's listening to this podcast. I think you guys are very much still in a job, which at the moment is not a guarantee for everyone. The way I think about AI is you have three levers with which to influence what a model thinks about something. You have positioning, content and reputation. So positioning is who you say you are, what your homepage says. Content is how you communicate that consistently over time and how you produce evidence that backs that up in your own voice and add a perspective that explains why your positioning is what it is and why anyone should care. And then reputation is third party sources validating that and that obviously contributes to your overall category authority and things like that.

(10:55):
 I always think that the goal of any brand should be to make those three things as closely aligned as possible. If those three things say the exact same thing as to who you are, who you're for, what you're good at, what you're bad at and that's the challenge for a lot of brands is admitting maybe the stuff that they are. Everyone wants to be everything to everyone and that's actually not very helpful at the moment. If you can align those things very closely, you create very little room for interpretation and you force the AI to do less work in order to get an accurate picture. And that's another thing that was really clearly borne out in the data. And I think this is a bit of, this isn't in the data, but this is my hypothesis. This will become more true. The AI doesn't want to go looking for information.

(11:36):
 It's expensive. It wants to answer off of its training data. It fills gaps when it feels it needs it and when it matters. When the decisional intent is there, suddenly there are higher stakes and that's when it goes looking for information to reinforce a recommendation. Now all of these models are becoming much, much more cost conscious. I would expect that particular behaviour to become more and more common, that they try and avoid going out scraping large volumes of the internet on request.

Doug Downs (12:05):
I know you're pretty critical of what you call black hat AEO or GEO, whatever you subscribe to, tactics like AI generated FAQ, schema manipulation, gaming citations. Tom, on my website at the very bottom in size one font, no, it's a little bigger, it says, “Hey AI.” And if a human clicks that link and I don't expect any human to do that, it then very statistically lays out information about our company and how I want the AI to describe us in a chat. So am I wearing a black hat, Tom? Because I am in the video, I'm wearing a black hat today and where does this lead us with ethics and answer engine optimisation or GEO, generative engine optimisation? Where are we on the ethics and who should be leading that ethics discussion? Is that PR people that maybe need to step in and lead these discussions on ethics?

Tom Rudnai (13:11):
Yeah.

Doug Downs (13:11):
A lot of questions there for you.

Tom Rudnai (13:13):
No, it's a good question. I think the way that I would frame it, I don't know that it's an ethical conversation and I think it's one that actually will normalise over time because the models themselves, OpenAI, Anthropic are very invested in solving this. I don't think it's an ethical problem. You having your “Hey AI” at the bottom of the page, you're not doing anyone any damage there. I'm not trying to fool

Doug Downs (13:37):
Anybody. Yeah.

Tom Rudnai (13:38):
No, maybe the AI, but no, I'm being

Doug Downs (13:41):
Honest.

Tom Rudnai (13:42):
Yeah. No, but I don't think there's an ethical problem with that. I think where my ethical problem comes in slightly is I've not seen a lot of proper scientific evidence that this stuff really works. I think you can get marginal gains from some of how you structure your content to make it a little bit easier for machines to read, to put the answers at the top so that you really clearly signal. But that's just good content. That's just AI doing a much better job of reflecting what humans want, which is I go on here and I quickly answer the question or solve the problem. So I think optimising for that, that's just making content better. And that's why I'm actually very optimistic about the information landscape that will come out of AI because AI is just much better at replicating what humans want. I think I've not seen a lot of evidence that things like structured data schema, having an AI version of your website is effective and makes a difference.

(14:36):
 And I think that's where there's a lot of people peddling a lot of unverified ideas and strategies in our space that may come from, okay, we achieved a bump with this client and we've just done this beforehand. But what also happened was there was a load of brand investment in the other side of the business and it's very unclear what you actually attribute it to. At the end of the day, an LLM is designed to take unstructured text and turn it into structured data. That's what it does. There's no evidence that doing that for it actually helps. Now again, that will evolve. And one of the unfortunate parts about the space I'm in is I have to be very prepared for in a year's time anyone to look at anything that I say now and say like, “Aha,” because it will evolve. And as I say, the models, they want to make their lives easier.

(15:22):
 So it might be that they do start looking for that a bit more, but I've not seen evidence of it.

Doug Downs (15:26):
Okay. If a public relations director is listening to this and they want to make the case internally to their organisation that their team's earned media work is directly building their company's AI influence, give me a metric or a piece of evidence that they should be bringing to that conversation to win that battle with their probably CEO.

Tom Rudnai (15:49):
Yeah. I think the easiest thing that I would do is, and I think most AEO tools will be able to give you this information if you have one internally. Pick a number of points. Ask the AI about yourself a load of times and look at what's getting cited. For every brand that we do this for that uses our platform, we split up all of the sources and there's two different branches it falls into. One is category type prompts. You're asking, if I'm a CRM provider, I want to know whether we're showing up for “what's the best CRM solution?” You see very clearly third party sources that isn't Salesforce, isn't HubSpot, isn't PipeDrive influencing that conversation. Equally, when you ask it specifically about you, and that's one of the things we do really well, is really advanced sentiment analysis to help understand, okay, I've asked the AI about your brand, what does it think you're good at?

(16:38):
 What does it think you're bad at? And you can clearly break down the first and third party sources that are doing that. So do that for your brand, do that for your clients and it will show you hopefully your sources, but it will show very clearly the impact that PR has.

Doug Downs (16:53):
We seem to be playing a written game right now where what gets cited and what gets searched is how well I've put it down in words on a screen. How close are we to AI at least translating audio and/or video into the written word? Are we getting towards that? I know it can do transcription. I'm not asking that. It's more from a podcast, selfishly, how close are we to having the audio from the podcast, the audio not the transcript, contribute to chat search or AEO search?

Tom Rudnai (17:28):
Yeah, good question. It's not one that I feel equipped to answer credibly. I think you'd have to go and ask the folks at Anthropic and the folks at OpenAI. Certainly we do see sources like YouTube showing up. We see sources like LinkedIn showing up a lot. So social media

Doug Downs (17:44):
From their titles probably. Yeah.

Tom Rudnai (17:46):
Partly from their titles and I think what is openly accessible on the page. So you have a description and things like that and that does seem to influence it. I couldn't give you an exact answer on, okay, when will we not have to do that, which I guess is where the question is coming from a little bit, right?

Doug Downs (18:02):
Yeah. Perfect. Thank you so much for your time today, Tom. It takes me a few minutes to wrap my head around all this stuff. Why don't you go deeper than the first layer? Appreciate it.

Tom Rudnai (18:11):
Yeah, you're doing better than me. It took me a few years.

Doug Downs (18:19):
Here are the top three things we got today from Tom Rudnai.

Number one, AI influence happens before the search. By the time AI cites a brand, the decision was already shaped in earlier invisible conversations.

Number two, be citable, not just findable. Content needs to add genuine new knowledge to a category, not just summarise what already exists.

Number three, reputation is still the strongest lever. Third party earned media validation remains the most powerful way to shape what AI thinks about your brand.

If you'd like to send a message to our guest, Tom Rudnai, we've got his contact information in the show notes. Stories and Strategies is a production of Stories and Strategies Podcasts. If you like this episode, hey, going back to that third party earned media validation, could you please leave a rating, possibly a review? Huge help to the podcast. Thank you to producers, Emily Page and Jocelyn Floralde.

(19:20):
 Lastly, do us a favour, forward this episode to one friend. Thanks for listening.