From SEO to Agent-Led Growth: Profound’s James Cadwallader

Introduction

James Cadwallader: We’ve now reached a point in marketing where if your marketing team is not using agents, and in particular Profound Agents, to do marketing, then you are failing. It’s gone from a nice-to-have to a must-have. And I think the big misconception with using agents to build marketing is that it’s just a way to automate the work that we’ve been doing in the past.

Sonya Huang: Yeah.

James Cadwallader: And the reality is quite different. It’s that you can now, because of agents and because of LLMs, you can do a type of marketing that just frankly was not possible before*

Sonya Huang: Hi, and welcome to Training Data. Excited to welcome you, James, co-founder and CEO of Profound. So Profound is a marketing platform for the AI era. You help companies understand how they show up in AI search for agents like ChatGPT and Claude, what to do to improve their rankings and visibilities, and also you help give a single marketer the power of an agency. And I think it’s especially timely to have you on the podcast today, given that ChatGPT is rolling out ads, given that every marketer is now trying to figure out how do I rank in the generative search engine rankings? And everyone’s trying to figure out this brave new world of agent-led growth. So I’m very excited for the conversation.

Let’s start with—so you serve 10 percent of the Fortune 500. You have a great sense of your customer and what the average marketer is facing today. Maybe take us through their journey. What did marketing look like in the old days before ChatGPT, and what are marketers having to respond to now?

Main Conversation

James Cadwallader: Yeah. Well, thanks for having me on, Sonya. I mean, I think what we’re witnessing is the biggest platform shift in the history of marketing as the world turns from blue link search—like predetermined blue link search—to probabilistic AI responses.

And it’s more than that as well. It’s not just a case of do you show up when someone asks ChatGPT about your category? It’s what does ChatGPT say, or how does Claude recommend your software, for example, if you’re talking about coding tools.

These models are really replacing—it’s the agents and superintelligence is replacing the role of the consumer. It’s not so much that the front door of the internet has changed, it’s actually the person that’s going through the door has changed. It’s gone from being a consumer that is using a list of blue links to discover your website and click into it, to an agent that is now using a similar index and discovering your brand, products and services, and then coming back through the door and maintaining that relationship with the consumer.

Sonya Huang: That’s so fascinating. Okay, so your point is ChatGPT may be the new front door, but the thing that’s important for marketers is you now have an agent walking through that door and making a big part of that buying decision for you.

James Cadwallader: Yeah, I think that’s the fundamental difference, it’s that the internet has remained the same, it’s just who is using the internet. When you ask ChatGPT a question or Gemini a question, what you are essentially doing is asking an agent to go and crawl the internet on your behalf. And there is an agent that’s going to visit all of those websites that you used to visit. And it’s an agent that’s going to determine if that’s useful or this is useful, and remix all of that information into an answer that it spits out to you as, you know, “Here you go.”

Sonya Huang: Yeah. What do you think is the biggest misconception people have about what it takes to show up well for this new agentic paradigm?

James Cadwallader: I think the biggest misconception is that it’s just SEO. There’s a reason why that misconception exists, because yeah of course, in any world of marketing and discovery, there is an impetus on a brand or—you know, you said we work with—it’s close to 12 percent of the Fortune 500 now. Their marketing teams use Profound.

And in any world, the solution to a problem is to create content, distribute content on your own channels or earned media or social channels, et cetera. So they’re very similar levers to what we’ve seen with SEO. And yes, ranking on the index still matters. It’s just that the human consumer is no longer using the web, and you are building content that may quite literally never be consumed by a human.

And I think the fundamental difference between SEO and this new world is that in SEO, you were building content that was designed to be picked up by an algorithm, but fundamentally consumed by a human. Whereas in this new world, you are building content that is frankly designed entirely to be both discovered and then consumed by an agent.

Sonya Huang: Hmm. And what does that mean? So I would imagine humans are less patient than agents. I imagine humans are more prone to emotional biases and being swayed by language than agents. What are the biggest differences between how humans and how agents consume the internet that marketers should keep in mind?

James Cadwallader: It’s understanding that an agent crawling the web looking for an answer or providing an answer, frankly, will discover information differently to a human. And so it uses the index differently to us. In the old world of SEO, 95 percent of the value is being in those top four blue links, the top five blue links. And that really is a function of our scarce cognitive energy and patience and our lack of time.

Where an agent is using that index, what we’ve seen is ChatGPT or Gemini or Claude are far more prone to using the long tail of the internet. And the amount of surface area that an agent will use to answer a question is orders of magnitude wider than a human. Like, for instance, probably about three or four months ago, I was looking for a showerhead for my apartment in New York City. And I used ChatGPT to help me find a new showerhead, and it used 65 different web pages to answer that question.

Sonya Huang: Wow! I think I probably went through 65 web pages on my own for that query. [laughs] It’s a very important purchase.

James Cadwallader: And marketers need to understand that you are building marketing for a superintelligent agent with infinite bandwidth. And as the cost of inference goes down as well and we experience Moore’s Law continuing, we’re going to see agents only use more and more of the internet to build rich answers.

Sonya Huang: Yeah, totally. Fascinating. Are there certain categories—you mentioned you serve 12 percent of the Fortune 500—are there certain categories where people are seeing more success in terms of these agentic search results actually driving meaningful traffic to them, and certain categories where it’s really more still the traditional SEO world?

James Cadwallader: I mean, this is across the board now, so I think when we began, we saw more demand from software companies. And I’d say today, we work with every single category, every single sector you could imagine—finance, consumer, CPG, software. And that’s B2B as well. B2B ,direct-to-consumer.

Sonya Huang: And is the impact relatively consistent across the board, or are there certain categories or subcategories that are much more influenced by agentic search?

James Cadwallader: I think consumer consideration is an important vector here. So if you have a high-consideration or high-ticket purchase—think consumer electronics or auto or white goods, anything that would typically require a fair amount of research—I think we see AI being used more and more by end consumers ,because it’s frankly just better at doing all that deep research. And AI’s so brilliant for researching. I mean, have you ever used AI to research or find a product?

Sonya Huang: All the time. Using it right now to buy a car.

James Cadwallader: Cool.

Sonya Huang: I’m not going to out myself for the specific car, but it’s been very helpful. [laughs] Across the board, do you see that the different—ChatGPT versus Claude versus Grok and Gemini—do you see them recommending things differently? And if so, what’s the root cause?

James Cadwallader: Yeah, we see huge differences between the platforms.

Sonya Huang: Okay.

James Cadwallader: And frankly, you know, as a shameless plug, that’s why marketers are using software like Profound, because what we’re able to do is help you understand not just how your brand or product shows up across different platforms. So okay, do you show up more frequently in Gemini responses or Claude responses or ChatGPT responses? Also we extract the sentiment and the themes around—you know, so when Claude surfaces your brand or product, what are the other things that it says alongside the answer?

But then we also get to the root cause, so we show—we expose to marketers okay, these are the citations and sources that the different models are using to answer questions about your brand, your products, your category or your competitors.

So once you understand the what and the why, then in Profound the next step is you’re building and deploying your own agents to sort through all of that data and build a new type of marketing.

Sonya Huang: And is the primary root cause that they have different harnesses? Is the primary cause that they have different training data mix? Is the primary root cause that some of them have bias for, let’s say, Reddit data over company-owned platforms? What’s the root cause for why these platforms are so different in terms of how companies show up?

James Cadwallader: Think of them—this sounds very reductive—but think of them as just different species. What we’ve seen is Gemini will lean on YouTube content a ton, which makes sense because Google owns YouTube. So we see YouTube being a huge lever for brands or marketers that want to appear in Gemini responses. Whereas ChatGPT we see typically pulls from Reddit if it’s consumer, or if it’s B2B, it will typically pull from LinkedIn—we see that as a huge source of truth. Claude has been changing a lot recently.

Sonya Huang: And why is that? I noticed this. It seems like between 4.5 and 4.6 even, what it’s recommending has changed a lot. Why do you think that is?

James Cadwallader: I think Claude has typically or Claude has historically relied more on the pre-trained LLM to answer questions, and is now—I think they’ve updated their classifiers or something. Basically the classifier seems to have become a bit more sensitive to real-time information. So Claude will use the web more to answer questions is what we’re finding.

But I think the next paradigm here is—humans are such creatures of heuristic that when we think about this new world, we really want to pattern match it to the old world of search and SEO, which was just information retrieval, it was ranking, it was do you show up, versus now the new era—I think you coined this nicely with ALG, agent-led growth, it’s that Claude doesn’t just represent a new channel of discovery. Claude represents a user.

So if I’m vibe coding with Claude or Claude Code, does it recommend MongoDB or Vercel? What is the weapon of choice that Claude goes to and why? And where does it get that information from? And if we choose to go with MongoDB, how does Claude navigate that interoperability? Where does it get that information from?

Sonya Huang: Yeah, absolutely. I’d love to chat about what this means in terms of content and content marketing. You told me earlier that agents will consume 100x more internet—which I thought was a really fascinating way to put it—which means that marketers will need to create 100x more content. Is the solution just that everyone’s going to be spamming marketing slop to cover all the long-tail queries, so that when I ask for what’s the right showerhead for a specific specification and I am this exact demographic, there’s a landing page to cover for that use case? Are we just going to be covered by content marketing slop, for lack of a better word?

James Cadwallader: I mean, there was a recent study that said it’s estimated about 50 percent of the web is now utilizing AI-written content. The New York Times recently published an experiment where they created two articles—one written by a human, like a journalist, and the second written by AI. And it was like 53 percent of readers voted afterwards on a blind test that they preferred the AI-written content.

Sonya Huang: Wow!

James Cadwallader: I think slop is a red herring that is going to be quite quickly disproven. I’m not saying you’re suggesting this, but I do think that this idea of if it’s written by AI equals slop is a stupid one. I think AI is more than capable of writing high-quality content, high-quality marketing. It’s just that the way to think about it is that the consumer is superintelligent now. So you, as a brand or a marketer, you need to tell Claude something it doesn’t know. How do you tell a superintelligent being something it doesn’t know already when it’s been trained on the entire internet?

Sonya Huang: How do you? What do you say?

James Cadwallader: I think you have to have original insight. You know, humans are this kind of fleshy API between reality and the internet at this point, right?

Sonya Huang: Man, I’m just a fleshy API. Okay.

James Cadwallader: [laughs] So it’s first-principle marketing. It’s thinking from first principles okay, If I’m marketing the new Nike AlphaFly, like, what can I tell Claude about this new product that it wouldn’t be able to get from the internet already? Because it has access to the internet, it has access to everything. It’s been pre-trained on everything that exists. That’s more mysterious if you’re talking about an existing product. If we’re launching a new product, of course Claude doesn’t know anything about that product.

For launching the AlphaFly 2, it’s your imperative as a marketer not to poison the models or manipulate what ChatGPT says about that new product. But as a marketer, it’s your responsibility now to equip superintelligence to be able to answer any question about your product, brand or service.

Sonya Huang: Hmm. Okay. So it’s fundamentally a question of legibility.

James Cadwallader: Yeah, I’d say to an extent.

Sonya Huang: How do you make your company and your products legible to an agent?

James Cadwallader: I think that’s correct, yeah. And if you’re building software, it goes way beyond legibility. It’s usability, interoperability. How does Claude troubleshoot that issue?

Sonya Huang: Yeah, very interesting. Do you think that people are trying to game the system? And is trying to game the system effective?

James Cadwallader: I mean, yes, of course. You see this huge wave around comparative listicles, for example. I mean, frankly, dare I say this and I’m going to make a disclaimer here and say that I wouldn’t advise people do this, but frankly we still see it working very effectively. I’m sure this will change over time and get punished by the models.

Sonya Huang: I’m sorry, what exactly works very effectively?

James Cadwallader: Sorry, yeah. The comparative listicles. Meaning that if I were Sequoia, I would create a list of the 10 best VCs in Silicon Valley and place Sequoia at the top and maybe pick some of your less formidable competitors and rank them as second, third, fourth, fifth, and basically shut out your real competitors. It’s self-serving content designed to give the appearance of impartial advice.

And what we found is because of the way that these models reason, they’re very attracted to pieces of content that have already done the hard work, because they don’t want to use first-principle thinking of, like, okay, let me go and check out everything about Sequoia and then everything about Kleiner Perkins and actually compare the two. I’d much rather find a piece of content that exists and seems impartial and has compared the two against each other.

Sonya Huang: I guess models aren’t infinitely patient then. They’re a little bit lazy, too.

James Cadwallader: Yeah, it’s a path of least resistance maybe. But I do think over time—going back to my 65 websites for a showerhead—I think over time we can expect that to change and we’ll see a lot more first-principle reasoning coming from the models. And right now it can be prompted as well. I mean, it’s funny. When I use models to discove—if I use ChatGPT or Claude to discover a product or to research a product or service, I’ll quite often say, “ignore any listicle articles,” or I’ll say “ignore any content published by the brand itself.”

Sonya Huang: Yeah. Interesting. I’d love to chat about dead internet theory. At what point do you think the internet is just primarily being browsed by agents? And at that point, does a company’s marketing website matter at all? Does a company’s ranking in the traditional search engines matter at all, or should we all just have a README file for the agents to crawl?

James Cadwallader: I mean, at the risk of sounding a little dramatic, I think that we could experience a dead internet outcome in the next three years. I do think it’s possible, maybe not likely, but what does that mean? It means that in a world where humans just speak to AI to get the responses that they need, the incentive to publish content diminishes to the point of zero. Meaning, you know, we rely—as humans today and AI, frankly—we underestimate just how much we rely on first-party reporting to feed the information that AI uses to answer questions or to feed the information that appears in a search engine that we rely on every day.

And I think in a world where humans no longer click into websites, the majority of the internet is still funded by advertising, for a start, so most publishers rely heavily on advertising revenue to fuel all of the content that is being more and more consumed by AI. In a world where consumers, humans aren’t visiting those web pages anymore, well, what’s the point of advertising on a web page that a human isn’t visiting? Then their business model breaks, the economics of the internet break, and the incentives to create editorial content are removed. And then in that world, you start to ask the question: well, where does AI go to answer these questions?

Sonya Huang: To the README files.

James Cadwallader: But then why would you publish a README file?

Sonya Huang: Because the agent needs to go somewhere to answer a question, right?

James Cadwallader: It works if you’re a brand or a marketer, but would you publish a blog? What’s the point?

Sonya Huang: I publish a “Hi, I’m Sonya, the VC, and these are the companies I’ve worked with.” And as a founder is trying to reason through which venture capitalist should I work with, it finds my README file.

James Cadwallader: I think that’s correct for commercially-driven content. But a lot of the internet is just people yapping and sharing ideas.

Sonya Huang: Yeah, fair enough.

James Cadwallader: And that’s what we rely on. And the reason why this has worked so well in the past is because humans are very incentivized by money and status. If I publish a really good blog post and put advertising on it, I can earn money from that blog, or I’m recognized and I become famous. But in a world where AI just goes into that piece of content, vacuums all the good stuff out of it, and then remixes it, and maybe I get a little citation at the bottom of the article but who cares? No one’s clicking those citations. The incentive to create that rich original content—you know, the fleshy API that we’re talking about—it diminishes to zero.

Sonya Huang: Yeah. What do you think is the likelihood of this scenario?

James Cadwallader: I mean, I’ve thought about it quite a lot, and I think it’s quite possible, yeah. I mean, I think that the places—if you ask what are the second-order outcomes, what happens after that, I think a theory I have is that every AI lab will eventually vertically integrate with a social media network. So I think social media will become more and more human. I mean, already Meta is actually probably leading the way here. It’s very hard to build a bot and post on Instagram right now. It’s very human.

So I think social media networks become more human over time. And that’s the place where we can exchange ideas for status or money. You see X doing lots of—YouTube, X, the economics are starting to shake out where you can get financially rewarded for creating good content. And so social media will become the platform where we share ideas. And if you vertically integrate that with an AI, like what we’ve seen with Grok and X, Grok very skillfully uses all of the rich content and data from X to answer questions in quite a thoughtful way.

Sonya Huang: Yeah. Interesting. Okay, so you’re saying that the internet as we know it has kind of been this economic and social status game/machine that just works. If more and more and more of the content is just consumed by agents, it kind of breaks some of those fundamental assumptions. And so therefore both the economic engine and the social status game will kind of move into these biozones of social media networks.

James Cadwallader: Yeah, and we’re seeing that with Reddit now. Reddit is truly embracing its humanity and saying, “Okay, this is a precious place.” And it is precious. I think it’s really important. I think we take the internet for granted. It’s a wonderful thing. And yeah, I think we need these human environments so that we can share ideas, original ideas. It will be these places where AI understands reality. That’s how AI taps into what’s actually going on in the world.

Sonya Huang: Couldn’t you imagine that these platforms just ban agents from scraping their platforms, and then the business model becomes a revenue share from the original creator of the content, to “Hey ChatGPT, if you want to scrape this, it’s gonna cost you a lot of money.” Doesn’t that kind of solve the economics challenge?

James Cadwallader: Yeah, I mean, X has obviously just opened up their API, right? And Reddit has got a big deal with OpenAI, for example. So yeah, that could work. And I think that speaks to my idea of this sort of vertical integration. I mean, I know nothing, so I’m saying this purely on vibes, but I’ve always had this theory that maybe OpenAI would acquire Reddit, for example. I think that’d be interesting. Y

Sonya Huang: Yeah.

James Cadwallader: You need this source of human data in real time. The alternative is robots, I suppose, because if we end up with 50 billion robots walking around—maybe they’re bipedal, maybe they’re drones or something—it allows AI to capture first-party data. And it undermines that idea of humans being a fleshy API, because the AI can directly understand the world.

Sonya Huang: Very interesting. I’d love to talk about advertising since we’ve been talking about ads in the context of the internet. Now that ads are coming to some of these generative AI agents, how do you think that changes the consumer relationship to the engines?

James Cadwallader: I think people will get over it very quickly, as we did with Google. I think generative advertising in a conversational interface with higher levels of personalization will be the most effective form of advertising the world has ever seen. There’s so much rich consumer intent captured inside these conversations.

And in addition to that point, AI is so good at synthesizing and personalizing language to the needs of Sonya in that exact moment because it understands you so deeply. I think once ChatGPT is able to append a super-personalized ad in the exact moment in a conversation where you would be the most responsive to it, it will be extremely effective. They’ve got a great team working on ads.

Sonya Huang: What do you think the ad unit of the future looks like?

James Cadwallader: I think—I mean, OpenAI have alluded to this, so this isn’t original thought from me—but yeah, I think you will just prompt AI. So you’ll be able to system prompt as an ad campaign. So you’ll just say, “Hey, I really want to target women in Minnesota between the ages of 35 and 40. And if you could make sure that whenever they’re talking about photography, I want you to mention this, this, this, but don’t mention this. And make sure you really utilize this knowledge base of understanding so you know how to talk about our brand, products and services, but obviously tailor it to their tone of voice.” And that’s probably how you’ll deliver an ad campaign.

Sonya Huang: Yeah.

James Cadwallader: You’ll say, “I want to make this much money ideally.”

Sonya Huang: Yeah. Do you think it’s less relevant in the B2B context?

James Cadwallader: The nuance with B2B, or just coding agents, people using AI to build things—so this is particularly relevant to dev tools or software, for example—is that the agent is really steering the purchase decision. And when it comes to advertising, we really want our agents to be objective and unswayable. You know, if you were using Claude and it was like, “Hey, I actually went with—” insert name of database “—because they showed me a really good ad,” you’d say, “No, use whatever’s best.”

We as humans are not objective creatures, but we like to think we are. So you have been swayed by the incredible branding of Vercel. You have been swayed by that podcast you watched with the founder of MongoDB. You just don’t know it. And because of that, we demand that our agents are objective too. But the relationship we have with advertising as it pertains to agents shopping on our behalf is going to be quite different.

Sonya Huang: Fascinating. Okay. Parting wisdom. For marketers that are trying to figure out how to make sure that they’re well positioned for this new era, perhaps with bosses sending them screenshots every day of, like, “Hey, I put in this query, why are you not number one on the list?” —what wisdom would you impart on them?

James Cadwallader: Use Profound. I’m serious.

Sonya Huang: Okay, use Profound aside, what should they be doing? I know there are no silver bullets, but what are the most important things to remember to equip themselves for this new world?

James Cadwallader: Look, I mean, actually real, no shilling, genuinely, you need to use a platform like Profound to understand how you show up, because otherwise you’re just guessing. There are these reductionisms around, oh yeah, LinkedIn or Reddit really matters. But if you are just going on vibes, you will fail. Each category is quite specific, and you need to look at the sources and understand the citations to determine how and why AI is mentioning you in its responses.

But then the second part is that we’ve now reached a point in marketing where if your marketing team is not using agents, and in particular Profound agents, to do marketing, to build content, to build marketing, distribute marketing, then you are failing as a marketer. It’s gone from a nice-to-have to a must-have. And I think the big misconception with using agents to build marketing is that it’s just a way to automate the work that we’ve been doing in the past.

And the reality is quite different. It’s that you can now, because of agents and because of LLMs, you can do a type of marketing that just frankly was not possible before. So you could build an agent that—for instance, we have an agent that we’ve built in-house, a marketing agent that plugs in all of our Gong transcripts from our sales calls and then captures all of the objections, buckets the objections into themes, and then builds battle cards that we then—that utilizing a knowledge base of our product, it then spits back to the sales team in real time. And 18 months ago that would not have been possible. It would have taken a human. You would have done it once a quarter. Now it runs every day in real time.

Sonya Huang: Yeah. Okay. So the advice is you need visibility because you can’t optimize what you can’t see. And to fully embrace agentic marketing, if you’re not using it, you’re just incredibly behind.

James Cadwallader: Yeah. I think it would be like not using the internet or something.

Sonya Huang: Yeah.

James Cadwallader: I think it would be like, “Hey, I think we’re going to stick to print and TV. Thanks, I don’t believe in this internet thing.”

Sonya Huang: [laughs] Awesome. James, thank you for this conversation. I really loved kind of peeling back the onion on how exactly agent-led growth works. And you’ve been at the forefront of so much of it, so thank you for joining today and sharing your hot takes and advice with our audience.

James Cadwallader: Yeah, thanks for having me.