The Rise of Dark AI and the Cost of Getting It Wrong

Mark Evans: Hi, it's Mark Evans and you're listening to Marketing Spark. Today's conversation sits at the intersection of something every SaaS company is trying to figure out how AI is changing the way the buyers discover, evaluate and choose solutions. And there's a growing focus on AEO as evidenced by AI-generated answers, citations and visibility in tools like ChatGBT, Gemini and Clon. That's where most of the attention is going.

But what if that's only a small part of what's actually happening? My guest today, Tom Rudd and I, has been looking at what he calls dark AI. The idea that a large part of influence is happening before brands are ever cited in the early conversations where buyers are framing the problem and narrowing their options. His team's research suggests that most interactions don't include citations at all, which raises a different question.

Buyers are making decisions before you're ever mentioned. Where should you be focusing? Thomas is CEO of Demand Genius. And today we're going to unpack what's behind this idea, what the data says, and what it means for how SaaS companies think about positioning, content, and growths.

Welcome to Marketing Spark. Thank you for having me. I'm looking forward to the conversation. Why don't we start off with a softball question.

When you talk about dark AI, What's the simplest way to understand what's happening beneath the surface? We like to think of AI as a new form of search or an extension of search. That's two point. When you actually look under the hood at how these models behave and what they do, it's not at all.

They actually very rarely search for an answer. They compile answers, they converge on answers. And when you understand that, it really changes how you look at the channel because If an AI isn't going out for searching for answers consistently, then the way that he went from site gets very different. It also means that it's not able to site it.

We tend to focus on citations, share a voice, things like that. But the AI doesn't want to mention. Search relied on the people that it sent it and it needed those links. AI doesn't.

It only does so when it is really necessary for the user, i.e. when the user is specifically asking to be directed to someone. Now when you think about a complex bite journey, Very few of those interactions do you want to be directed to a vendor.

If somebody is speaking, you want to get a question answered and help frame your problem, define your requirements, all of those things. We know in B2B that's the majority of the channel. That's nothing. So that's what we mean by dark AI.

It's all of those little interactions from day one of being problem-able through to actually looking for solutions where AI isn't searching for answers, it's not mentioning you. AI is influencing the buyer's requirements. is influencing the way that the problem is threatened and we know that all of that adds our ultimate to the decision we're making. Much more than just who is visible at the point of convert.

I think what you've highlighted, the fact that we're barely scratching the surface when it comes to AI and our appreciation of how it works and how we should use it. On a related note, when you think about how many companies, whether they're B2B or SaaS, are approaching AI and the level of sophistication they have when it comes to understanding how the models work and how their marketing and sales can leverage the power of AI and these models. Where do you think they're at? If you were a CEO and you saw this AI tsunami in front of you, would you be confident right now that your team knows what they need to do?

Or do you think there's a giant learning curve that a lot of CEOs need to put their companies on? not. Just talking to marketers, I'm always struck by it. I think they really feel like they're in a vice and I'm getting away from AEO and more just AI in general.

AI is a very interesting trend because it's defined very quickly by marketing dollars and not more than actual case stuff and use cases and validated examples of working with and working with practice. It's also very difficult for a marketer because the earliest adopters and empower users initially of AI are investors and CEOs and founders and all of the people who impose expectations on you. So you're caught between Platforms that over promise under deliver over promise to your boss and then you're left holding a can or holding the bag when it under delivers.

You don't have any real, the demands are no less of you. If anything, they're higher in terms of quality and quantity. But these tools often they're quite the never. You're the one who with ever decreasing resources, good content is now commod.

Code is commod. Has to deliver on those things. So it's a difficult time. What does that conversation look like between the CEO and the head of marketing and the head of sales?

The CEO obviously has these high expectations. They recognize that AI could be a game changer for the business. Marketers are mandated to embrace AI in some way, or form because it's the way that they can become more efficient and maybe in the process, reduce headcount or at least hold headcount. So what does that conversation look like?

What is the CEO, what should the CEO say to the marketer? What should the marketer say back to the CEO? It's different in every organization. It depends a lot on the size of the organization and on the CEO and the marketer.

What a lot of marketers feel it's manifesting at the moment as a crunch on a head count. It's do more with less or really like it's being used as an efficiency tool to try and cut up humans. So I think we will continue to go on that flow for a little bit until eventually people realize that actually the best results from AI come when you mix human insight in really intelligently and you use it as a tool not to automate what humans could already do. but to actually unlock complete new possibilities.

I'll give you an example. Our content that did, that's actually about the darkening I thought that we're talking about. Right. That's something that a startup of our size could not have attempted five years ago.

We AI to capture an immense amount of data that was qualitative in nature. Right. So that's what AI does really well. The qualitative analysis, subjective analysis makes judgment calls at scale.

So you simply couldn't have conducted that analysis five years ago. And that's what led us to do it. were like, actually, how can we use AI to do something on a scale that we couldn't have done before? Really big, that rather than just to churn out lots of low quality pieces of content.

And that's, think the journey that everyone is then going to go on is once we compress head count and we realized that we did get lower results as we dropped those things, they'd fall off together. We then start building head count back up, but we build it up in a much smarter way and much more emanated. To the benefits. Yeah, it's fascinating.

think we're in a period of learning. We're in a period of transition. I think a lot of people are just trying to figure out how they should move forward. And that is an immense challenge right now.

Thank you for referencing the AI report that, that your company published earlier this year in the report on dark AI and AEO influence. What specifically did you see in the data that convinced you that influences are reset before citations appear rather than citations being one of the several factors shaping the final decisions. I guess to load up the question even further, what were your surprising findings from the report? Everything surprised us because we honestly didn't go into it with too much of a kind of set belief.

I had an inkling of a hypothesis that every, all of the advice out there that AEO is just SEO. like, seems incorrect. That's very different pieces of technology. It's so hard to give us the exact same best practices for how to optimize for them both.

But I had a little hope that something seemed a little bit off there. But then when we got into it, think I was very surprised at how stark the trend was in terms of citations. So what we saw, and for context for anyone listening, what we did. We ran an immense number of prompts and we wanted to look specifically at how AI responses vary across complex B2B biojourneys.

we 14 complex categories, different levels of complexity, all B2B. And we ran prompts very deliberately across awareness, consideration and conversion phases. And then we looked at, we analyzed the response to our technology and our AI to understand how they vary. And what we saw was, look at citation?

Or actually let's start that. Mention, you get mentioned. at every stage of the funnel, but inevitably the more bottom of funnel you get, the more you get through to your conversion rate increases pretty dramatically. You look at how often retrieval is invoked.

This is what really surprised me actually. And when we talk about it not really being a search channel at all in awareness, 0 % of the time retrieval was invoked. In consideration, 0 % of the time retrieval was invoked. In conversion, then it was 48%.

For the vast majority, we know that the bulk sits in a innocent consideration and it doesn't go looking for answers. So your content, it's not useless because over time you can influence training data, but it's a pretty long horizon. That it's not a very well understood thing. And accordingly with that citation rate was very simple as well, zero zero 48%.

So all of our measurement, all of our strategy at the moment is about the idea that I produce a piece of content and tomorrow it shows up in AI directly cited. Simply not. how these models work. half of those Blendit file queries is that a trend that is plausible and there's a whole load of stuff we can go into for like it still isn't quite that direct.

If a marketer came to you or a CEO came to you and said, Tom, what are the three takeaways from your report? What should I do? How should I act upon what you've discovered from this in-depth piece of research? What would you tell them?

So I'm going to break it into kind of measurement and then strategy. I think for measurement, have to develop new metrics that are fit for the new problem rather than trying to apply it all over, right? SEO style metrics. The best one that we've been able to come up with, and this is new, right?

So I don't have all of the answers. I'm always very keen to say that. It is looking at fit. So there's an immense different number of ways that I can ask a single prompt to AI.

Right? So if you just try one of those and measure the response and say that I get Cytos, there's so many different variations. What is very predictive of how you are represented across all of those different interactions, every stage of the journey, just tells you what does AI think of you. So how does it view your strengths, your weaknesses, your fit for different buying criteria, different use cases.

You can map that to different personas that you want to influence and get a really good sense of, okay, not so this query that it Cytos, but for a CFO. But a big company who we know wants X, Y, and Z, does it think we're good ⁓ or not? And if the answer is not, what does it say we're bad at and how can over time we're still correct? And that produces like, if you can make sure that whenever the CFO at your big deal you're trying to close goes on and asks, you good value for money?

You don't want to make sure it says yes. And that's more valuable than any amount of visibility at the top of the funnel. So that's why we're measuring what if you focus on fit more than visibility. demonstrating that the right use cases.

how do you define SIT? What's the lamest version of SIT? So you can quantify that. We run sentiment analysis on how your brand is represented for different criteria.

So let's take, if I'm buying, we'll go with a commonplace example. I'm buying trade, which is relatively high consideration purchase. There's different buying criteria that people might apply. They will come from they want.

spigginess, they want them to be environmentally friendly. They want them to have a really cool switch up aside. Whatever the criteria people might buy, we can ask the AI, how good do you think Nike, Adidas, all of these are for that? Run sentiment analysis on that.

Then that gives you your really clear breakdown of what it thinks you're good at and bad. And then you go through a process of mapping that to what your different segments or your different stakeholders might want. I won't get into the maths to underlie it. That is something that you can quantum.

and trackers of the sentence in a KPR. So that's number one. Number two piece of advice would be what? From the reports.

Stop viewing content as a one-to-one route into a particular query. Take a step back, search at its very most basic level. The job that you do with search, you find a really high intent query, ⁓ a very common or high intent query, high volume or high intent. You write a piece of content for it that was summarizing knowledge against that query and the best summary one.

That's search in a very simple nutshell. AI doesn't work like that. You would need to find there's like an endless number of those queries. What it looks at more is the entire corpus of your content.

Because it doesn't, when I put in what CRM should I buy, doesn't go searching for an answer to that. That one piece of content, which you keyword stuff with what CRM should buy, isn't doing you any good. What is doing you good is the perception that you build up. over all of your content as to who you are, who you're for, what you're good at, what you're bad at.

So that's quite a big, that's actually quite a big shift because it changes the approach to content creation because a lot more of your effort then has to go away from pumping out new content and towards maintaining improving content that you already have. want all of your content to be high quality, consistent, clear. and current, right? If it's all of those things, it's consistently painting the same picture of your brand to AI and to humans.

As a content marketer, I'm not sure whether I should be terrified of that point of view or excited. The pressure not to create content means I don't know how much my job is going forward. On the other hand, the focus on high quality content and making sure that the content I already published is optimized is actually not a bad thing because if I'm proud of what I've done, obviously I I'm incentivized to make it better. If you're a marketer and you're going to say to your CEO, this is our approach to content marketing.

You have to not look at it as single bullets that we're firing to see if we can hit the target. It's going to be almost like a slow burn. The totality of our content and the quality of our content, that will make the difference in terms of AI. So we need to take a more pragmatic, more measured.

more quality focused view of the world. that the way that I should interpret it? Yes, I think so. would say as well, like bearing in mind, whatever, I know it's XYZ is dead.

It's the very common refrain on like, like search is not dead. There is still SEO. I'm talking about your SEO and that there is space for both of them. But you can blend that in a way that achieves the results you need in the timeline you have and stuff like that.

But strictly from an AER perspective, yeah, it's quality over quantity. Every piece of, we have this kind of framework called content debt. Every piece of content you put out, you take on a small amount of content debt because the amount of monthly quarterly work that you have to do just to maintain your overall library just increases ever so slightly. So when you understand that, you can be very deliberate about how you do it.

We're an early stage company. We're going to stop putting out. content are actually relatively high volumes. We're good with how we do that and we have this kind of core quality threshold so that we're really careful.

But we need to build authority in our chosen areas. And so we actually need some volume. And so you potentially, it's like building technology, you accept the time, it will take on a bit of that. You just have to have an understanding of, know that it has to be.

One of the other sort of analogies that you talk about is the idea of an iceberg. And curious about. the mechanism underneath the metaphor. Like how does that, how does a brand become viable before it's ever mentioned?

listeners, the metaphor also for me, it's an epiphany that I don't know if it's actually as visual for other people as it was, but basically the search for the funnel again, because all of the volume sits at the top and then you try and push people down. is the exact inverse, right? All of the, your, you get your traffic from the bottom of funnel posts, not the top of funnel posts, which means that you actually get a lot less traffic. So the way that we think about the iceberg is what above the surface is that 16 % of AI responses where there is a citation.

So they're the ones that under current metrics you can track. if you say we're, AI is not that important to us. We're only seeing this many citations and then we get a hundred hits a day. Well, you are seeing the tip of the iceberg is 16 % of from producer citation.

And then what percentage of those actually click through to your site. So if you're seeing a hundred hits, you can multiply that massively to see actual overall impact that AI is having. It's a very common, for anyone who's watched the Titanic, the vast majority of the maps of any iceberg sits beneath the surface. And that's just the visual that I find really helpful for understanding AI.

Anything that's showing up in your current metrics is the tip of the iceberg. So if your beads successful in terms of tip of the iceberg activity, that means that you're probably doing really well with activity below the surface as well. You just can't see it. I think it does and it doesn't.

What I can't, I can't tell you definitively yet the correlation between success at the tip and success beneath the surface. think the tip is, are you visible when someone searches for your solution? Everything below is, are you influencing problem framing and requirement building in a way which gets more people to search for your solution? It's like performance marketing versus brand marketing or positioning.

And that's what it all comes to. What I find interesting about that analogy, if in fact it's accurate, is that top of the funnel content, especially in B2B and SaaS is sexy. It's fun. It's lists.

It's 10 things you should know. It's why we're the best. It's the kind of stuff that is low-hanging fruit, easy to engage. It's the kind of stuff as marketers, we think that consumers like to read.

it's. I'm I'm using the wrong word, but it's Bottom of the funnel content is dense and heavy and big on product education and showing people how platforms work. And as a marketer, I can tell you it's tough slogging. It's important, but it's the kind of, it's like going to the gym and lifting weights.

You have to do it, but it's really isn't that exciting as opposed to being on the elliptical where you're racing really fast. It's kind of, it's leg day. Yeah, it's a leg day. That's what it is.

That's your spryers. Ship dingers a little bit. When you look at AEO and how companies should approach it, when you see how companies are trying to frame themselves as the only option, a lot of them lead to brand positioning and the idea that if you have clear and compelling brand positioning, that's how you're going to attract and influence buyers. But in the AEO world, influence is a different creature.

Can you provide some insight into... that traditional mindset of ⁓ positioning versus the way that buyers behave and what they react to in this AEO world. Could you, could you expand on the question a little bit? In the traditional world, if like we all lead into brand positioning as what do you do?

Who do you serve? How are you different? And what's it for the buyer? And if we feel, if we get that message out into the marketplace, then that's going to be enough to make them say, wish I should check out brand XYZ.

But in the AEO world, I don't know if that stuff matters as much as opposed to creating enough content, enough high quality content. it compliment? good brand positioning or is it more important than good brand positioning? Like how do get that brand story out in an AEO world?

So there's a few things that we've seen work really well. think the first is Specificity is actually really important. You mentioned it, like the all-in-one platform. That's not a very good claim to make anymore.

way that's used to CRM category is what a lot of people know. If I'm a CRM startup right now, I'm not going to be the CRM for everyone, for everything. Sales, sports and house bought are probably going to have pretty locked down. What I can do that can go a lot more specific and going back to those use cases, I can say, okay, I'm the best CRM for small medium businesses in London who are really interested in MCP or AI first integrations or something cool like, after you.

Specificity is a really easy way to embed your positioning. And when the criteria that AI over the course of a long set of interactions with the user, what it does during that time is it gathers context and applies criteria. And if you can find your positioning, which is a big enough meaty part of it, then it's going to eventually come to you at that point. There's an analogy or an image that I quite like here, which is think of like a spotlight on the street.

You can run around at the bottom, try to like, it's very difficult to hover. If you're trying to get into the spotlight, you have a couple of things that you can do. You can run around chasing. That's what I most people are doing at the moment.

Or you can actually go further up towards it. And then suddenly it's an awful lot of ease because you can, the spotlight either grows in a percentage. So that's what you're trying to do is go up towards it. and try and influence the way that AI frames your category for specific groups of people in a way that will ultimately make visibility inevitable rather than you having to chase it.

So how do you do that? The biggest thing that we've seen is original research, but I don't know if you saw Google did a big core update recently in one of the biggest trends that came out of it, people have analyzed that is 22 % visibility boost in the AI overviews. for content that based on original research versus content that wasn't. So we have this concept called information gain, which I think should be a KPI for every content team out there at the moment, which is basically again, if we go back to that old role of content query, summary of knowledge against that query, best one wins.

That is useless now. If all you're doing is summarizing knowledge and get probably getting AI to do that for me on a bespoke basis. I don't need your content and AI doesn't need your content. It's useless to humans and it's useful to machines.

What is useful and what does get consistently picked up and reinforced with positioning is having some sense of your content IP or some kind of original research. That's what we call information. Something that produces net new knowledge that is worth citing, that is worth incorporating into how AI understands your category. That seems to really consistently boost the authority of your site and lead to influence.

We actually, have a framework for measuring, which is you have different tiers. So there's level zero, which is no information gain. Level one is interpretive gain. a new slant on an existing piece of knowledge, right?

And you can achieve that with AR if you're really good at how you see your positioning, your market perspective in and have it produce level one. Level two is empirical information gain. So new data, new research, survey, whatever it is. AI won't produce that.

And then level three is conceptual information. So backed by original research, something genuinely new and innovative. What I would like the dark AI is for us as a new concept. Right.

Picked up. We're going to go back to your question about a while back, what the CEO should do. One of the biggest things I would, is investing in that original research function because it feeds emphasis. Basically you're eating your own dog food by creating this original research and focusing on the dark AI.

What's happened since you published the report? Have you been able to see like within AEO or within SEO or within in-bounds? Is that kind of thing? Are you actually, is it working for you?

It's an interesting case study. You're telling other people to do, but is it working? Yes or no? Cause it's not instantaneous.

think so. And yes, it's produced. inbound and produce resolves and it allows me to go and have conversations like this. That has a powerful effect.

You get press coverage, things like that. It's helped to establish our new positioning. We pivoted into this program a little bit and it's helped to establish that within the models. If you go and ask care about us, we're not quite at a point yet where it's going to mention us over some of the competitors.

It's a good case study. That's one of our big priorities for the next couple of months is to solve that. A deeply hypocritical gap for us. What you have when you're early.

As I said off the top, I get a lot of in bounds for being guests on the podcast, just like probably every other podcast or these days in the days of automation, I would say the 95 % of them I project because they're irrelevant or off topic. But when I saw the pitch about dark AI and AEO, was like, well, I could rally around that. So it's banked by research. Yeah.

everyone wants. I think particularly in this space, is so much, it's a space being defined by content marketing rather than research. And we tried to do that for a little bit. then I was like, hold on, why don't we actually just use facts to cut through the noise?

That'd be a pretty good idea. People remember storytelling, but they don't remember facts. That's part of a storytelling presentation that we all marketers like to give. guess facts do matter.

If you're the CEO of a fast moving company, your marketers are probably using AI. I suspect they've got Claude or... chat, GPT, pro subscriptions. They say they're using it to do frameworks and to do research.

So they're probably using it to create content. Maybe it's not the highest value content, but in some way, shape or form, they're using it to create content. But if you want to be on the, on the cutting edge, if you really want to start to leverage AI and harness the power of AEO. Where would you start from the top down?

What kind of direction would you start to give to your teams, whether it's sales or marketing or customer service in terms of, got to figure out, we got to do this right. And we're to focus on the fundamentals and we got to have the right plan at the right time. What's your advice in terms of how a CEO would get started? think my first thing would be that you have to give people space to play and you have to give them permission to do that.

And that requires a little bit less obsession with efficiency. think it's sad fact of technology and AI to all create this hyper focus on how efficient someone is and it creates no space to play with that. So I don't think it's a coincidence. When I said earlier, the early adopters are CEOs and investors.

in control of their own time and can say, know what, going to spend today playing with Claude and I'm going to see what I can do. If you're in a team, you've got strict KDIs and someone breathing down your neck for what I can do. If I had someone doing that every day, the number of days where I'd have to say, I was playing with this, didn't work. And that's not acceptable.

So you have to create like the freedom and the culture that people can do that and look and judge the outputs of it over and long. time horizon and like, did you do today? Yeah, that's what's going to produce. The other thing I would say, challenge people to produce quality with it.

Make sure we could go in for ages to how you do that. I would challenge people in the team not to replace something, not to have it do something end to end. Look at a process. See, if I had 10 times as much time to do this thing, what would I do?

And how can I use AI to then go and achieve that? Because AI does have all the time in the world, all the patience in the world. And when you're really clever with how you blend it into a process that works, you can unlock so much. can improve the output so much rather than just multiply the output.

I love the idea of quality over quantity because in the AI world, quantity is easy. It's easy to scale your content, a couple of props and you've got 10 blog posts for the rest of the month, what do want to do? how, like CEOs are busy, they put other priorities and they're not always watching what marketing is doing. How do you make sure that there's alignment in terms strategically as far as Everybody knows where you need to go strategically, how you need to leverage this new wave of technology.

But how do you, how does the CEO ensure that best practices are being followed, that their efforts are being successful? Because at the end of the it's all about ROI. How are we going to spend our resources and what can our return on what we can get? What kind of tools or insight does the senior executive team need to have to make sure we're doing the right things?

It's difficult because I think it's really hard to put like a cop lodge rule because it's going to be different for every team. So I'm going to have that. I take us and on this particular topic, that's all I can really speak to other than people that I immediately know and obviously talk to about this. But for us, we're a relatively small team.

I use AI to create alignment a lot. So I've built a system where basically via MCP, all of my, this conversation, which I'm recording, I'll have my notes from Granola. My notes are one to one, will feed into a central document, which manages our good market alignment and the team in Claude can go and because it has my one to ones with everyone in the team, they can say, what's John up to? What's Steve up to?

Whoever the people are with the team and they can ask. So it keeps everyone aligned on a kind of project basis through that. And that helps because when you're asking lots of people to play, that becomes a really high risk that you just have five people playing with the same thing and trying to solve the same problem. So trying to consolidate all of that into a way that they can get quick feedback on whether an idea is worth pursuing before the next one-for-one, I think is really important.

To the best of your ability. Right. I think within a leadership team. Again, I think it's going to vary depending on where you are, but I think it would just be setting the expectation that you're using AI to solve problems and setting the, giving the kind of freedom to then fail to solve that problem.

I think is where people get tricked up. I love the idea of experimentation. spent a lot of my days experimenting with QOD, especially cowork as a game changing tool in terms of how I'm using AI. think that.

You're right that a lot of marketers are under pressure to perform. And that means that often they they're afraid of making mistakes or doing something that doesn't follow best practices or a campaign that isn't as successful as it should be. And maybe Google was on, on the right track when they said that 20 % of their time should be spent experimenting. think marketers need that.

And salespeople for that matter need that too, is that you, there's tons of tools out there. There's tons of things you can do and sometimes they're not going to work and that's okay. But great conversation, Tom, thanks for this. The idea of dark AI is super interesting.

The report, which I'll share in the show notes is a must read. some, there's a ton of really great research in there and it's good to see that you're walking the walk and talking the talk when it comes to that kind of thing. Where can people learn more about you and Demand Genius? Download our website at demand.

genius.com. research there and we publish it all really openly. So go to the bottom, we outline the full methodology by all means, go reproduce it.

But if you find something different, tell me, we'll publish that too. We're very committed to doing this with the market and figuring out this new thing that we all have to learn how to do together. So probably one of the few things that I promise you is that I will have a different story a year from now. If we do this again, I'd say I'd encourage you to do that and also connect with me on LinkedIn, Tom Rudnay.

Rudnay is very difficult to spell. R-U-D-N-A-I. I know it's happening. I should ask if I'm a B2B or SaaS, why should I call you?

What can you do for me? What's your raison d'etre? ⁓ God, is this the elevator? Hey, this is the elevator.

There's no pressure now. You've got to me some warning. I'll help you build an AEO strategy that focuses on influence over visibility. So see what AI thinks of you.

Understand how you can better align content positioning reputation to influence that over time and then track the revenue impact of that as well. pieces of content and that sort of influencing. Thanks Tom for the great and insightful conversation. And thanks to everybody for listening to another episode of Marketing Spark.

If you found this conversation valuable, subscribe on Apple podcasts, Spotify or your favorite podcast app. Drop a quick rating and share it on social media. You can reach me by email, mark at mark Evans.ca connect with me on LinkedIn or visit marketingspark.

co. I'll talk to you next time.