James Cadwallader | Training Data Summary

1. Key Themes

The Fundamental Shift: Agents Are Now the Consumer, Not Just the Search Engine

The most important reframe in this episode is that the internet itself hasn't changed — who uses it has. Agents are now crawling, synthesizing, and making decisions on behalf of humans, which means the entire logic of marketing must be rebuilt from scratch.

"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 is now using a similar index and discovering your brand products and services." — James Cadwallader [00:02:36.140]

Agents Consume 100x More of the Internet Than Humans

Unlike humans, who are cognitively constrained to the top 4-5 blue links, agents traverse the long tail of the internet at massive scale. This completely changes the surface area marketers need to cover and the type of content they need to produce.

"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... ChatGPT used 65 different web pages to answer that question." — James Cadwallader [00:06:08.180]

The Dead Internet Theory — A Real 3-Year Risk

The economic engine of the internet (advertising-funded content) may collapse as human page visits drop to near zero. This is not a fringe theory — James gives it meaningful probability within three years, with significant second-order consequences for AI training data sourcing.

"I think that we could experience a dead internet outcome in the next three years... 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... the majority of the internet is still funded by advertising... but in a world where consumers, humans aren't visiting those web pages anymore, well, what's the point in advertising on a web page that a human isn't visiting?" — James Cadwallader [00:18:39.760]

2. Contrarian Perspectives

AI-Written Content Is Not Slop — It May Actually Be Better

Against the prevailing narrative that AI-generated content degrades quality, James argues the opposite. A blind NYT study showed readers preferred AI-written content 53% of the time. The real problem isn't quality — it's originality.

"I think slop is a red herring that is going to be quite quickly disproven... this idea of if it's written by AI equals slop is a stupid one. AI is more than capable of writing high quality content, high quality marketing." — James Cadwallader [00:13:40.500]

Every AI Lab Will Vertically Integrate With a Social Media Network

This is a bold structural prediction: as the open web degrades, the only reliable source of real-time human signal will be social platforms. AI labs will need to own or deeply partner with these platforms to stay grounded in reality.

"I think a theory I have is that every AI lab will eventually vertically integrate with a social media network... Grok very skillfully uses all of the rich content and data from X to answer questions in quite a thoughtful way." — James Cadwallader [00:21:25.740]

"I've always had this theory that maybe OpenAI would acquire Reddit, for example. I think that'd be interesting. You need this source of human data in real time." — James Cadwallader [00:24:00.820]

Generative AI Advertising Will Be the Most Effective Form of Advertising Ever Created

Counterintuitively, ads in conversational AI interfaces — despite seeming intrusive — may outperform all prior formats due to unprecedented personalization and intent capture.

"I think advertising or generative advertising, I should say, 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 of these conversations." — James Cadwallader [00:25:18.680]

Comparative Listicles Still Work — and Brands Are Weaponizing Them

Despite the assumption that AI models are sophisticated enough to detect self-serving content, manipulative listicles ranking competitors favorably for yourself are currently still highly effective at influencing AI responses.

"Comparative listicles... if I were Sequoia, I would create '10 best VCs in Silicon Valley' and place Sequoia at the top... self-serving content designed to give impartial advice... 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." — James Cadwallader [00:16:27.100]

B2B Advertising in AI Is Fundamentally Broken — Agents Won't Accept Influence

In B2B and developer tool contexts, the agent makes the purchase decision, not the human. Humans demand their agents be objective, which means traditional advertising persuasion mechanics simply don't apply.

"We as humans are not objective creatures, but we like to think we are... 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." — James Cadwallader [00:27:29.820]

3. Companies Identified

Profound

Description: AI-era marketing intelligence and execution platform
Why mentioned: Serves 12% of Fortune 500 marketers; enables brands to monitor visibility across AI platforms (ChatGPT, Claude, Gemini, Grok), understand citations and sentiment, and deploy agents for content creation and distribution
Quote: "What we're able to do is help you understand not just how your brand or product shows up across different platforms... but also we extract the sentiment and the themes... and then we get to the root cause." — James Cadwallader [00:09:05.780]

Reddit

Description: Human-generated social discussion platform
Why mentioned: Identified as a critical source of truth for ChatGPT's B2C responses; has a major data deal with OpenAI; positioned as a rare "precious" human environment that AI depends on
Quote: "Reddit is truly embracing its humanity and saying, okay, we, this is a precious place... it will be these places where AI understands reality. That's how AI taps into what's actually going on in the world." — James Cadwallader [00:23:12.760]

Grok / X

Description: Elon Musk's AI model integrated with X (formerly Twitter)
Why mentioned: Cited as the leading proof point of vertical integration between a social network and an AI model — using human social data to produce richer answers
Quote: "Grok very skillfully uses all of the rich content and data from X to answer questions in quite a thoughtful way." — James Cadwallader [00:22:19.980]

4. People Identified

James Cadwallader

Description: Co-founder and CEO of Profound
Why mentioned: Built a platform now used by 12% of Fortune 500 marketers; has a deep, data-grounded understanding of how different AI models source and rank information; coined or helped popularize the "agent-led growth" framing
Quote: "We've now reached a point in marketing where if your marketing team is not using agents... to do marketing, then you are failing. It's gone from a nice to have to a must have." — James Cadwallader [00:29:27.200]

5. Operating Insights

Build a Gong-to-Battle-Card Agent Pipeline Now

Profound built an internal agent that ingests Gong sales call transcripts daily, clusters objections into themes, and auto-generates battle cards grounded in product knowledge — a workflow that used to take a human a full quarter. This is immediately replicable for any sales-driven organization.

"We have an agent that we've built in-house, a marketing agent that plumbs 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... it runs every day in real time." — James Cadwallader [00:29:47.200]

Platform-Specific Content Strategy Is Now a Requirement

Different AI platforms pull from fundamentally different sources. YouTube dominates Gemini. Reddit and LinkedIn dominate ChatGPT. Claude is shifting toward real-time web. One content strategy no longer fits all.

"Gemini will lean on YouTube content a ton... ChatGPT 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." — James Cadwallader [00:10:53.760]

Marketers Must Now Create Content for Agents, Not Humans

The entire mental model of content marketing — write for humans, optimize for algorithms — is obsolete. The new imperative is to make your brand and products legible, interoperable, and navigable for a super-intelligent agent with infinite bandwidth.

"You are building content that is frankly designed entirely to be both discovered and then consumed by an agent... it's your responsibility now to equip super intelligence to be able to answer any question about your product, brand or service." — James Cadwallader [00:05:17.440]

6. Overlooked Insights

The Ad Unit of the Future Is a System Prompt, Not a Creative

This was mentioned briefly and neither speaker dwelt on it, but it represents a massive structural shift in the advertising industry. The entire creative, media buying, and targeting apparatus of modern advertising may collapse into a single system prompt. This has profound implications for ad tech infrastructure, agencies, and brand strategy.

"You'll build a system prompt as an ad campaign... hey, if you could make sure whenever they're talking about photography, I want you to mention this... tailor it to their tone of voice. And that's probably how you'll deliver an ad campaign." — James Cadwallader [00:26:08.360]

Claude's Classifier Update Is a Stealth Platform Shift

James casually mentioned that Claude changed its underlying classifier between versions 4.5 and 4.6, making it significantly more sensitive to real-time web information. This kind of silent model-level change can overnight invalidate months of brand visibility work — and almost no marketers are monitoring for it. This creates both a risk and a competitive intelligence opportunity for companies tracking AI visibility.

"Claude has historically relied more on the pre-trained LLM to answer questions and is now... 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." — James Cadwallader [00:11:15.660]