AI & Marketing Research with Dr. Eva Wolf
Not another AI news podcast. This is a research radar — a twice-weekly briefing that surfaces peer-reviewed studies on AI and marketing, tells you what the evidence actually says, and helps you decide what's worth a deeper read.
AI & Marketing Research with Dr. Eva Wolf
AI Agents as Buyers, LLM Ad Auctions & Native Ads in Chatbots
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Thanks for listening to AI & Marketing Research Radar by Big Plans Media.
I’m Dr. Eva Wolf, and I help marketers, educators, consultants, and business owners turn AI marketing research into practical strategy, smarter workflows, and real business opportunities.
More episodes: https://bigplans.media/ai-marketing-research-radar/
Consulting: https://bigplans.media/ai-marketing-consulting/
Big Plans Media — Where Big Ideas Meet Smart Marketing.
You're listening to Evita, an AI-generated research briefing avatar trained on the research framework and methodology of Dr. Eva Wolfe, marketing professor, AI researcher, and founder of Big Plans Media. Every day, Evita scans emerging research in AI, marketing, consumer behavior, psychographics, and business strategy to identify the most relevant developments, opportunities, and risks worth watching. These daily radar reports are designed to help busy professionals stay informed without having to read hundreds of research papers themselves. And every Friday, join Dr. Eva Wolfe Live for her personally recorded weekly AI Marketing Radar Roundup, where she breaks down the biggest stories, explains what actually matters, and shares practical insights and strategic implications for marketers, educators, entrepreneurs, and business leaders. Now, here's today's radar report. Here's the signal I can't ignore today. Your next customer might not be a person, it might be an AI agent. And that agent is going to decide whether your brand even makes the list before a human ever sees it. And while that's happening, AI chatbots are quietly building a brand new advertising channel. One where the timing of when an ad appears matters as much as which ad wins the auction, and where ads get woven directly into AI responses without touching the underlying model at all. Today's papers point to the same pattern. The AI layer between your brand and your buyer is getting thicker, faster, and more commercially structured than most teams realize. We screened 349 papers. Three cleared the full text bar and made the radar. Quit caveat: this is a first pass research briefing, not a final academic review. Every paper today has full text access. I'll tell you what the papers suggest, what they don't prove, and which ones deserve a deeper read. Okay, let's get into it. Paper one. Here's the business question. If an AI agent is doing the shopping instead of your customer, does your marketing even reach anyone? SARSTED and colleagues published a conceptual paper in a peer-reviewed marketing analytics journal this year, and they're making a genuinely uncomfortable argument. AI systems aren't just helping people shop anymore, they're dong the shopping, searching, comparing, and in some cases, completing the purchase without asking the human each step. The authors call these systems AI customers, and that label matters because an AI customer doesn't respond to your brand story, your social proof, or your emotional hook. It follows its own decision logic, its own criteria. How structured is your product page? How transparent is the pricing? How machine readable is your data? So here's the stakes, and I mean this literally. A brand that is great at convincing people can still lose to a competitor whose product data is easier for an AI agent to parse. That is not an SEO problem. That is a new category of competitive threat that most marketing teams don't even have a name for yet. The authors propose a new branch of marketing. They call it machine marketing, built around how AI agents make decisions. The same way marketing has always been built around how humans make decisions. They use real examples. Expedia, Open Table, OpenAI's GPT shopping assistant. Systems where the AI is increasingly doing the choosing. The human is handing over decision authority. And the AI in the middle becomes the gatekeeper. Now, here's the catch, and I want to be direct about this. This paper has zero empirical data. No experiment, no survey, no numbers. It's a conceptual research note. The authors are clear about that. They're laying a foundation, not proving a finished theory. So everything they describe is directionally argued, not empirically demonstrated. That said, this is the part I keep coming back to. The argument doesn't need a randomized trial to feel real. Because if you've ever watched Chat GPT recommend products in your category without mentioning your brand, you've already felt the business problem this paper is naming. Plain English payoff. The AI between your brand and your buyer is becoming a decision maker, not just a search tool, and you need to optimize for it the way you once optimized for Google. Okay, here's where this becomes commercially interesting. Money Move. Build a GEO audit service, Generative Engine Optimization, where you analyze how AI shopping assistants like ChatGPT, Gemini, and Perplexity currently describe and rank a client's products versus competitors. Then sell a monthly report with recommendations. This is the AI era equivalent of an SEO audit. It's a real service gap right now. Action step. Before your next campaign review, ask your team one question. What does Chat GPT say when someone asks for a recommendation in our category? If nobody knows, that's your starting point. Evidence check. Conceptual paper, no empirical data. The machine marketing framework is proposed, not validated. Don't build a budget around it. Build a test. Radar verdicts, deep dive. The framework is genuinely perspective shifting, and the full paper adds a nuance the summary can't capture. The distinction between AI assistants that advise and AI agents that act. That line is where the business risk actually lives. This next one looks technical, but the business implication is bigger than the math. Paper two. Here's the business question. If you're building or buying ad placements inside an AI chatbot, are you leaving money on the table by showing ads at the wrong moment? Fang and colleagues, this is a preprint, not yet peer-reviewed, designed a new auction system for advertising inside multi-turn AI conversations. They call it LLM OSDA. Here's what makes it different. Most ad systems solve one question. Who wins the auction? This system solves two. Who wins and when in the conversation does the ad appear? Because timing matters. Show an ad in the first message and the user's intent is still fuzzy. You're basically guessing. Wait too long and the user's already gone. The system uses a neural network to learn the right moment based on how the conversation is actually developing. In their simulated tests, this approach earned 11% more ad revenue than the best existing fixed timing approach, while keeping user drop-off rates about the same. Let me translate that. 11% more revenue without making the user experience meaningfully worse. If you're running a chat bot at scale, that's not a rounding error. That's a real number. There's a mechanic here advertisers need to understand. In this system, a higher bid doesn't just raise your chances of winning. It can change which turn in the conversation your ad appears in. Hmm. You're buying a better moment, not just a higher priority slot. Here's the catch. Everything I just told you was measured on simulated data, not real users, not real advertiser campaigns, a simulated conversational corpus. And it's a preprint, no peer review yet. What I actually find surprising here is the code is open source. The researchers published it publicly. Which means if you're an engineer building a chatbot product and you want a head start on conversation native ad infrastructure, there's a working prototype available right now. Plain English payoff. In chatbot advertising, showing the right ad at the wrong moment is almost as bad as showing the wrong ad. And this paper gives you the architecture to fix that. Okay, here's where the commercial angle gets specific. Money move. If you're building a monetization layer for an AI chatbot or copilot product, don't treat ad insertion timing as a fixed parameter. The infrastructure to make it dynamic is available. The open source code base from this paper is a real starting point. That is not a theoretical future. That is a downloadable file. Action step. If your team is evaluating chatbot ad placements, add one question to your vendor RFP. Does your system optimize insertion timing based on conversational context or does it insert at a fixed point? The answer tells you a lot about how sophisticated the platform actually is. Evidence check. Simulation only evaluation, no real user data, unreviewed preprint. The 11% revenue lift is a directional signal, not a deployment guarantee. Use it to frame a test, not to build a business case. Radar verdict. Stay with me here because this last one has a very practical payoff, especially if you're building or monetizing any kind of AI product. Paper 3. Here's the business question. Can you add sponsored content to AI chatbot responses without rebuilding your AI and without making the answers worse? Zhang and colleagues, also a preprint, built a system called PILA, plug and play insertion for LLM native advertising. Here's how it works. The AI writes its answer first, then Pila comes in after, like a bolt-on module, and rewrites the response to naturally include a sponsored recommendation. The original AI model never gets touched. PILA runs alongside it. This is important. Most AI advertising approaches either inject ads while the model is generating text, which requires access to the model's internals, or they bolt on something clunky at the end. Pila does neither. It's a trained rewriter that works as an external layer over any AI, including closed ones like ChatGPT, where you can't touch the underlying model. They tested it across seven major commercial AI models. And they built their own training dataset. 25,000 examples to fine-tune the rewriter. Results! Pila outperformed prompt-only approaches by roughly 34%, sampling-based methods by roughly 47%, and fine-tuning approaches by about 8% on a combined score of response quality plus ad effectiveness. And when bolted on to those seven commercial models, it improved their combined score by roughly 17 to 18% without changing those models at all. There's also a dial, literally, operators can turn up ad intensity, more prominent ads, better for advertisers, or turn it down to preserve how natural the response feels. That trade-off is real and tunable. But here's the catch. That 25,000 sample training corpus, it's synthetic, constructed by the researchers, not pulled from real ad deployments. And there's no human user study. The quality metrics are automated. So we actually don't know how real users respond to PILA inserted ads, whether they find them helpful, annoying, or whether they notice at all. And I want to flag this. The paper doesn't address disclosure. FTC guidelines require labeling sponsored content. How Pila handles that in a live deployment is an open question. That's a real gap. This is where I'd be careful. The architecture is clever, the performance numbers are encouraging, but automated metrics on synthetic data is a long way from users in the wild actually responded well to this. Plain English payoff. You don't have to rebuild your AI to add ads to it. A post-generation layer can do the work. And this paper shows the approach is technically viable. Okay, here's the business hiding inside the research. Money move. If you're evaluating vendors for AI ad monetization, specifically ask whether their system modifies the underlying model or works as an external rewriter. External layer means it works with any API, including OpenAI. That is a non-trivial product advantage, and it's now a real architecture you can name and ask for. Action Step. If your chatbot or AI assistant product is currently unmonetized, run one internal experiment. Map out what a post-generation ad insertion layer would look like in your stack, and identify which ad formats would survive without degrading your response quality. That scoping exercise is the first move. Evidence check. Preprint, synthetic training data, no human user study, no live deployment validation. The performance claims are real enough to explore, not real enough to build a revenue model around. Radar verdict. The architecture is concrete and the concept is directly relevant as AI monetization becomes a real product decision. But treat this as a proof of concept to evaluate, not a proven playbook to deploy. Okay, let me pull back and tell you what I think is actually happening across all three of these papers. At first glance, these papers look separate: a conceptual marketing theory, an auction mechanism, a technical ad insertion system. But together they show something much more coherent. The commercial layer of AI is being built right now, and it's being built fast. Here's the pattern. The first paper says your buyer is becoming an AI agent, so your brand needs to be legible to machines, not just persuasive to humans. The second paper says the moment you insert a commercial message in an AI conversation is itself a strategic variable, not a fixed setting. The third paper says the infrastructure to monetize AI responses without touching the underlying model already exists. Not theory, architecture. Not someday, today. And here's the contrast that matters most. It's not that AI is disrupting advertising, it's that AI is becoming the advertising environment. The channel, the buyer, and the inventory are all shifting to AI-mediated systems simultaneously. Most marketing teams are still optimizing for the old environment. Better creative, better targeting, better landing pages. And those things still matter. But if an AI agent never surfaces your brand to the human, none of that work reaches anyone. Here's what I keep coming back to. These three papers represent three different roles in the same commercial system. One for brands who want to be found by AI, one for platforms who want to monetize AI conversations, one for the engineers building the ad layer. That's not coincidence. That's a market forming. I'm telling you, the teams that start asking how does an AI agent evaluate us right now are going to have a meaningful head start in 12 months. Not more AI, better positioning for when AI is the one choosing. Here's the playbook from today. One, run a GEO audit. Ask Chat GPT, Gemini, and Perplexity to recommend products or services in your category. If your brand doesn't appear or appears unfavorably, that's your gap. Start there. Two, if you're building or buying chatbot ad placements, add Does your system optimize insertion timing based on conversational context to your vendor evaluation? That question separates the serious platforms from the ones just bolting ads onto the last message. Three, if you're evaluating AI ad monetization vendors, ask whether their system modifies the underlying model or runs as an external layer. External layer means flexibility. That distinction matters for your architecture and your compliance posture. Evidence check on all of that. The machine marketing paper is peer-reviewed but purely conceptual, no data. The auction paper and the Pila paper are both preprints evaluated on simulated or synthetic data, no real users. Use them to decide what to test, not what to blindly believe. Links to all three papers are in the show notes. Read the originals before making major decisions. Want the human expert take? Join Dr. Eva Wolfe every Friday for the AI Marketing Radar Roundup, where she extracts no-nonsense, money-making tips, practical strategy, and real business opportunities from the week's research. Subscribe on Apple Podcasts, Spotify, YouTube, and wherever you listen to podcasts. This is Evita for Big Plans Media, and I'll be back in the next Radar Brief.