AI Ad Targeting Research: LLMs, Knowledge Graphs & Ad Auctions

``` === AI MARKETING RESEARCH RADAR === === TODAY'S RADAR QUESTION === AI is already inside your ad stack — but are platforms using it as a scalpel or a sledgehammer? This episode digs into three new studies asking whether LLMs actually improve ad targeting in production, what happens when too many brands fight for control of an AI's recommendations, and how wiring a knowledge graph into your ad engine could make it both smarter and 24% faster. === PAPERS COVERED === 1. Fine-Tuned LLM as a Co -- 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

0:12

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 question nobody in digital advertising is asking loudly enough yet. When AI chatbots start running ads, and they will, who decides which ads get shown? And does it have to wreck the experience for users? That's the thread running through today's papers. LLM Ad Auction Mechanics, a 15-second video ad machine. And what actually makes AI-generated Facebook ads work or completely backfire. We screened 346 papers. Three cleared the full text bar and made the radar today. Quick caveat: this is a first pass research briefing, not a final academic review. Every paper covered 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 the first one. Paper one. The mechanics of how ads get embedded inside AI chatbot answers are being figured out right now, and most marketers have no idea this conversation is even happening. Two researchers built a new auction system for how ads get placed inside LLM-generated responses. Think Google AdWords, but instead of a search results page, the ad lives inside the AI's actual answer. Here's the problem they're solving. Right now, if you shove ads into AI answers, platforms just pick whoever bid the most, highest bidder wins, and what happens? The AI starts giving you irrelevant answers because the winning ad has nothing to do with what you ask. Answer quality tanks. Users get annoyed. Everyone loses eventually. So these researchers design something different. A quality filter. An ad has to clear a minimum relevance threshold before it even enters the auction. If your ad isn't contextually appropriate for this answer, you don't get to bid. You're out before the auction starts. And here's the part I find genuinely interesting. The new system actually earned more money per ad shown, not less. The ads that cleared the relevance filter were worth more because they fit the context. Advertisers bid honestly because the rules made it pointless to game the system. Higher revenue, more relevant ads. AI answers that stayed close to what the model would have said with no ads at all. You don't have to choose between making money and keeping users happy. That's the actual claim. And the math backs it up in simulation. Wait, that matters. In simulation. I want to hold on to that before I get too excited. This isn't a live platform test. It's a model. But the framework is clean and the logic is tight, which is more than you can say for a lot of what gets published. Plain English payoff. The platforms that win in LLM advertising will filter for relevance first and auction second. And the brands that understand this will get better placements at lower effective cost. Money Move. When AI ad platforms launch and Chat GPT, Gemini, Perplexity, they're all moving this direction. Push for placement in relevance-gated systems, not just highest bidder systems. Your click-through rates will be meaningfully higher because you're reaching someone already in the right context. Try this by Friday. Pull up whatever AI platforms you're watching. Perplexity, ChatGPT, Gemini. Spend 20 minutes reading how they currently handle sponsored content. You want to understand the rules of the game before the game officially starts. I'm telling you, the marketers who understand relevance-gated ad placement before it goes mainstream are going to have a real edge. Evidence check. This is a theoretical paper. No real users, no live platform, no actual advertisers. The whole thing runs on simulation. Treat it as a blueprint, not a proof. It's also a preprint, hasn't been peer-reviewed yet, so hold it loosely. Radar verdict! Read now. The mechanics of LLM advertising are being written as we speak. And this paper gives you the vocabulary and the mental model to understand what's coming before it arrives. Paper two, what if you could type your product name into a box and get a finished video ad? Three script options, rendered video, ready to download in 15 seconds. That's not a pitch. That's what these researchers actually built and tested. They built an end-to-end automated video ad platform. You put in a product name, the AI writes three scripts: one emotional, one feature-focused, one urgency-driven. You pick one, upload some photos, and the system renders a 30-second video. Total time from input to downloadable video, about 15 seconds, 3 seconds for the scripts, under 12 to render. Okay, here's the thing. They also ran a usability test, standard measure, system usability scale. Score came back at 84.6 out of 100. That's in the good to excellent range. Which means a non-technical user, think small business owner, Etsy seller, local service provider, could actually use this without a tutorial. And the three-script structure is smart, emotional, informational, urgency focused. Three distinct persuasion modes. You're not just getting one AI written script you have to accept or reject. You get real creative choice automatically. Now let me tell you where this falls apart, because it does fall apart a little. The video output is assembled from static product images, photos you upload. It's not generating original footage. So if your product photography is weak, your video looks weak. Garbage in, garbage out. And this is the part that bugs me, they never tested whether the ads actually worked. No click-through rate, no conversion data, no viewer engagement. They proved the tool is fast and usable. They did not prove the ads perform. Plain English payoff. AI can now write three complete video ad scripts and render a finished 30-second video in 15 seconds. Which means the bottleneck in small business video advertising just moved from production to creative judgment. Money Move. Automate the production, charge for the strategy, and the selection. That's where the margin lives. Try this by Friday. Use whatever AI video tool you have access to: Canva, Adobe Express, something similar. And generate three script variants for one of your current products. Emotional, informational, urgency. Compare them. You're not committing to production. You're just seeing what AI surface that you wouldn't have written yourself. Evidence check. The journal this was published in is low credibility. Self-reported impact factor, limited peer review transparency, and the usability sample size, the number of people who actually scored the tool isn't clearly reported. So that 84.6 score? Directionally useful. Statistically, hold it lightly. Radar verdict. Test this week. The engineering blueprint is real and buildable, but don't wait for academic validation on whether the ads perform. Run your own test. Paper three. So we've talked about how LLM ads will be structured and how AI can produce a video in 15 seconds. Now here's the consumer side of the equation. When people actually see AI-generated ads on Facebook, what happens in their head? Researchers surveyed 380 Facebook users after exposing them to both AI-generated and traditional ads. They wanted to know: does AI content drive more purchase intent? And if so, why? Here's the finding. And it's more nuanced than AI ads work. AI-generated ads boosted purchase intent, but not directly. The effect ran through perceived usefulness. When people found the AI ad helpful and relevant, they wanted to buy. When they didn't, when the ad felt off or irrelevant, the advantage disappeared. And here's the part that should make every meta advertiser sit up straight. Privacy concern acted like a break on the whole system. The more someone worried about how their data was being used, the less useful they found the AI ad, which in turn killed their purchase intent. The usefulness effect basically collapsed under privacy fear. So the AI ad advantage is not automatic, it's conditional. You get it when your targeting is right and the ad feels genuinely helpful. You lose it completely when the person feels surveilled. That's the piece I care about. Because most AI ad tools are being sold on the premise that personalization automatically wins. This study says, not if you're creeping people out. AI Facebook ads work when they feel useful and fail when they feel invasive, which means your targeting accuracy and your data transparency matter more than the AI-generated creative itself. Money move. Add a visible data transparency signal to your AI-targeted ad campaign. Something as simple as, we're showing you this because you searched for X, or a clear privacy preference link. This study found that reducing privacy anxiety is worth testing as a direct lever on conversion intent. Try this by Friday. Pull one of your current AI-targeted Facebook campaigns and look at it through the usefulness lens. Does the ad answer a real question the user was already asking? Does it surface something they were already looking for? Or is it just personalized looking without being actually helpful? That distinction between helpful and just targeted is exactly where the conversion gap lives. Evidence check. This is self-reported purchase intention, not actual purchase behavior. Big difference. And the 380 participants, we don't know where they're from, what their age range is, how they were recruited. Demographics are undisclosed. Treat this as a directional signal, not a definitive finding. This hasn't been replicated in a high credibility outlet. The mechanism it describes, usefulness as the engine, privacy fear as the brake, is plausible and consistent with other research, but it's not settled science. Radar verdict. The mental model is useful and the direction is clear, but the evidence base is thin enough that you should test the privacy transparency lever in your own campaigns rather than just taking the paper's word for it. Okay, so here's what I think is actually happening this week. All three papers are about the same underlying shift. AI is entering every layer of advertising. The auction that decides which ad runs, the tool that produces the ad, the consumer psychology that determines whether the ad lands. And the thread running through all of it is this AI advertising doesn't automatically perform better. It performs better when it's relevant, useful, and contextually appropriate. And it performs worse, sometimes dramatically worse, when any of those conditions break down. The LLM auction paper says relevance gated systems beat highest bidder system. The video ad paper says, AI can produce creative instantly, but usefulness still depends on what you feed it. The Facebook paper says AI personalization boosts intent only when the user finds the ad genuinely helpful, and privacy worry kills that effect entirely. Here's the tension I keep coming back to. The industry is moving fast toward AI-generated everything. Faster creative, automated placement, algorithmic targeting. And the research is quietly saying the speed doesn't matter if the relevance isn't there. You can have a 15-second video pipeline and still run ads that feel hollow. You can have the most sophisticated LLM auction and still lose users if the fit is wrong. The competitive edge isn't in who automates fastest. It's in who understands the relevance problem deeply enough to solve it at every layer. Here's the playbook from today. 1. Start learning how LLM ad placement works. Read about relevance gated advertising. When Chat GPT or Gemini launches a real ad product, you want to know what a relevance filter is and why it matters to your bid strategy. 2. Use an AI video tool this week. Generate three script variants for one product emotional, informational, urgency. See what surfaces. You're training your eye for what AI Creative does well and where it needs human judgment. 3. Audit one AI-targeted Facebook campaign for the usefulness test. Is the ad actually helpful or just personalized looking? If you can't answer that honestly, your targeting may be better than your creative, and that gap is costing you. Evidence check on all of that? Two of today's papers come from low credibility journals, and one is a preprint with no peer review yet. 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.