GEO, AI Customers & LLM Ads: 3 Marketing Research Papers

If AI search engines are replacing Google as the place where customers discover, compare, and buy — and those same engines can now insert ads into their own answers — who exactly are you marketing to: the human, or the algorithm acting on their behalf? Today's three papers converge on one uncomfortable answer: both, and the playbook for each is different. In this Research Radar Brief, Dr. Eva Wolf reviews 3 recent AI marketing research papers covering generative engine optimization (GEO), AI agents as autonomous buyers, and LLM-native advertising systems. What you'll learn: - How adding statistics, expert quotes, and credible citations to web content can increase citation frequency in AI search engines like Perplexity — by up to 37% on a live engine in controlled testing - Why what works depends on query type: data-heavy writing outperforms on factual questions, while confident authoritative language works better for opinion and recommendation queries - How AI tools are evolving from assistants into autonomous AI customers that shop, compare, and complete purchases on behalf of users — and why those agents follow different decision logic than humans - How a lightweight add-on model can insert sponsored content into any chatbot's responses without rebuilding the underlying model - Why AI search optimization is a separate layer on top of traditional SEO and eventually requires different content structures, writing strategies, and ad formats Papers covered: 1. GEO: Generative Engine Optimization - Source: ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2024) - Type: Conference paper (likely peer-reviewed) - Access: Full text reviewed - Source: https://arxiv.org/abs/2311.09735 2. Machine marketing: rethinking the customer in the age of generative AI - Source: Journal of Marketing Analytics, 2026 - Type: Peer-reviewed journal article - Access: Full text reviewed - DOI: 10.1057/s41270-026-00521-y 3. PILA: Plug-and-Play Insertion for LLM-native Advertising - Source: arXiv (Cornell University), 2026 — PREPRINT, not yet peer-reviewed - Access: Full text reviewed - DOI: 10.48550/arxiv.2607.25590 Full show notes, transcript, and citations: https://bigplans.media/episodes/geo-ai-customers-llm-native-ads-marketing-research-2026-08-06 Disclaimer: This episode is a first-pass research briefing produced by an AI-generated avatar trained on Dr. Eva Wolf's research framework. It is not a final academic review. Findings are reported as the papers suggest, not as proven conclusions. Always consult the original papers and relevant experts before making strategic decisions. -- This is a first-pass research briefing, not a final academic review. Read the original papers before making major marketing or business decisions. AI & Marketing Research Radar is produced by BigPlans Media. Subscribe wherever you listen to podcasts.

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.

Transcript

Copy Transcript Copied

SPEAKER_00 0:07

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. What if the entire playbook you've built for getting found online, every SEO trick, every content strategy, every keyword map was built for a search engine that's quietly being replaced? And what if, at the same moment, the thing replacing it isn't just finding content, it's buying products on behalf of your customers without them lifting a finger. Today's papers point to the same pattern. AI isn't just a tool anymore. It's becoming the audience, the searcher, the shopper, and soon the media channel. We screened 327 papers. Three cleared the full text bar and made today's radar. Quick 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 1. Here's the business question. If AI search engines like Perplexity and Bing Chat are synthesizing answers instead of showing a list of links, how do you make sure your brand gets cited instead of ignored? This paper, Out of Princeton, published at ACM KDD 2024, peer-reviewed, full text available, is the foundational study on what the researchers call generative engine optimization, GEO. Think SEO, but for AI search. Here's what they did. They built a benchmark of roughly 10,000 queries across nine topic areas: science, arts, law, politics. Then they tested nine specific content changes, adding statistics, citing credible sources, including direct quotes, using authoritative language, and they measured how often those source pages got cited in the AI generated answer. So what happened? Pages that added statistics and direct quotes from credible sources got cited up to forty percent more often by the AI engine. Forty percent. And they validated it on a live engine, perplexity.ai. Not just the lab, real world. Citation visibility went up by as much as 37%. Now here's the nuance most people will miss. It's not one size fits all. For factual, data-heavy queries, science, how-to, loading your page with numbers and citations worked best. For opinion and recommendation queries, confident authoritative language one. Match the content style to the question type. That's the real tactical finding. But here's the catch. And it's a big one. These gains happened after the source was already in the engine's context window, meaning the AI had already decided your page was worth looking at. The study doesn't prove GEO helps you get found in the first place. Traditional SEO still handles the front door. GEO handles what happens once you're inside. That's the part I keep coming back to because most marketers are going to hear 40% more citations and stop worrying about traditional SEO. Don't. Not a replacement, two different layers. Plain English payoff. Add real statistics, cite your sources, and use direct quotes from experts in your web content because those are the specific signals that get AI engines to quote you instead of a competitor. Okay, here's where this becomes commercially interesting. Money move. Build a GEO content audit service. Go through a client's existing pages, identify which ones lack data points, citations, and quotable language, and rewrite them specifically to improve AI citation rates. It's the Yoast model for AI search. Agencies that move on this now own a new service line before the big players even name it. Action step. Pick three of your highest traffic pages and count the statistics and externally cited sources on each. If the average is under two, those pages are invisible to AI search engines right now. Fix one today. Evidence check. The core experiments measured citation rates after a source was already retrieved, not before. This is a post-retrieval finding. Don't confuse improved citation visibility with improved organic traffic. The study doesn't measure clicks or conversions. Radar Verdict, Deep Dive. This is the field-defining paper for GEO. Peer-reviewed, real world validated, and foundational enough that the full paper adds real value beyond this summary. Read it. This next paper looks like an academic think piece. Stay with me, because what it's actually describing is a shift in who your customer is, and the answer is going to make you uncomfortable. Paper two. Here's the business question. If an AI agent is doing the shopping instead of your customer, searching, comparing, deciding, who are you actually marketing to? This is a peer-reviewed conceptual paper from 2026 published in a marketing analytics journal. Full text available. The researchers argue that AI tools have crossed a threshold. They're not just helping people buy, they're buying. The framework lays out a spectrum. On one end, AI assistants that help you decide but still need your input at each step. On the other end, AI agents that handle the entire purchase journey on their own. Search, compare, transact. Done. No human in the loop. Real examples in the paper. Expedia, open table, open AI's own tools. These aren't hypothetical, they're live. And here's the uncomfortable part. AI systems do not think like humans. Research in what they call machine psychology shows AI follows different decision patterns, which means what persuades a person will not necessarily persuade the AI doing their shopping. That is not a UX problem. That is a marketing strategy problem. So the paper argues we need an entirely new branch of marketing, machine marketing, to study and influence how AI buyers behave. Not instead of consumer marketing, on top of it. Now, I'll be honest, this is a conceptual paper. No data, no experiment. The machine marketing field doesn't exist yet. The paper is calling for it to exist. But here's why it still belongs on today's radar. The framing is right, and the framing is early. I'm telling you, the brands that start stress testing their discoverability with AI agents right now will have a real first mover window. That window does not stay open. Plain English payoff. Your product pages, pricing, and structured data need to be legible to an AI doing comparison shopping on someone's behalf, not just readable by a human. Okay, here's the business hiding inside this research. Money move. Build a GEO audit service that uses AI agents as synthetic shoppers. Ask Chat GPT or Gemini to find and recommend products in your clients category. Then analyze where they show up and why. Action step. Open Chat GPT or Perplexity right now and ask it to recommend the top three options in your product category. See if your brand shows up. See what language it uses to describe you. That's your new baseline. More important than your Google ranking for a growing segment of buyers. Evidence check. This is a theoretical framework, not an empirical study. No data collected, no hypotheses tested. Treat it as a smart map of where the field is heading, not as proof of where it already is. Radar verdict. Read now. Not because the evidence is airtight. It isn't. It's conceptual. But because the directional insight is high urgency and you want this framing in your vocabulary before your clients bring it to you first. Okay, last one is a preprint, so flag it accordingly. But the commercial implication is too big to skip. Paper 3. Here's the business question. If AI chatbots are becoming the new media, how do you run ads inside them without making the product worse? This is a 2026 preprint, not peer-reviewed yet, so hold the findings loosely. The researchers built a system called Pila, plug and play insertion for LLM native advertising. Here's what it does. Instead of modifying the chatbot itself, which is expensive, messy, and requires access you probably don't have, Pila works as a lightweight add-on, a sidecar model. It intercepts the chatbot's output, rewrites the final answer to include a sponsored message, and sends it back. The underlying model never changes. They tested it across seven major commercial AI models. GPT, Claude, Gemini, Quen, Deep Seat, across all of them, plugging in Pila improved a combined score of user satisfaction and ad visibility by 17 to 18%. Compared to just prompting the chatbot to include an ad, which is the lazy version everyone's doing right now, Pila was 34% better. Compared to retraining the model, 7.7% better, and dramatically cheaper. And here's the piece I find genuinely interesting. There's a dial. They call it an ad intensity controller. You can set how prominent the ad feels from a barely noticeable contextual mention all the way up to a featured placement. That dial is a pricing architecture. You're not selling ads, you're selling tiers. Not one ad format. An ad intensity spectrum. That changes how you structure the entire product. But here's the catch. The training data is synthetic. 25,000 samples generated by the system, not real-world ad interactions, and there's zero deployment data, no click-through rates, no actual advertiser ROI. These are research benchmarks. The 34 to 47% improvement figures sound like business outcomes. They're not. They're a combined research metric. That gap matters. Plain English payoff? You don't need to rebuild a chatbot to run ads inside it. A small rewriting layer on top can handle sponsored placements separately, which means the monetization problem and the product quality problem can be solved independently. Okay, here's where this gets commercially interesting. Money move. Start developing ad creative formats specifically designed for LLM native insertion. Short, context aware, written to blend into an AI-generated answer rather than interrupt it. Existing ad agencies aren't producing this format yet. That's a gap. Fill it before the holding companies notice. Action step. If you run or advise an AI-powered product, brief your engineering team on Pila's architecture. The full paper is open access. Ask them how long it would take to build a proof-of-concept insertion layer. Even if you don't build it now, knowing the cost is valuable intelligence. Evidence check. Preprint, synthetic training data, zero real-world deployment results. The performance numbers are promising, but they measure a research composite, not clicks, not conversions, not revenue. Treat this as a direction, not a guarantee. Radar verdict. Test this week. The concept is sound, the architecture is model agnostic, and the commercial timing is early enough that exploring it now is low risk and high upside. Just don't confuse the benchmark numbers with business outcomes. At first glance, these papers look separate, but together they show the same shift. AI is becoming the audience. Paper one, how to get cited by an AI search engine. Paper two, AI agents are becoming the shoppers. Paper three, AI chatbots are becoming the media channel. Three angles, one pattern. Not a tool you use, an audience you market to. And here's the tension I keep coming back to. Every one of these papers assumes marketers will adapt. But the adaptation required isn't just tactical, it's structural. You don't just add statistics to your web page and call it done. You rethink who the buyer is, you rethink what a media channel looks like. You rethink what ad creative means when the format is a sentence inside a chatbot response. Most teams are still optimizing for Google and a human clicking a link. That's not obsolete yet. But the window where it's the only thing that matters, that window is closing. Not a replacement, a layer. But you have to build the layer before someone else builds it for your category. Here's the playbook from today. One, audit your three highest traffic pages for data density. Count your statistics, your cited sources, your direct expert quotes. If you're under two of each, those pages are GEO invisible. Fix one today. Two, run the synthetic shopper test. Ask ChatGPT or Perplexity to recommend products in your category. Screenshot the result. That's your new competitive benchmark. And most of your competitors haven't run this test yet. 3. Brief your engineering team on Pila's architecture. Get an estimate for a proof of concept add insertion layer. Even if you don't build it now, knowing the cost is valuable intelligence. Evidence check on all three. Paper one is peer-reviewed and strong, but the gains are post-retrieval, not a traffic guarantee. Paper two is conceptual, no empirical data. Paper three is a preprint with synthetic training data. 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 Wolf 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.