AI Ads Inside AI Answers, Self-Evolving Ad Systems & Health AI

What if the ad slot you're bidding on today becomes irrelevant — not because clicks drop, but because AI generates the answer before users ever see a search result? And what if your customers are already using ChatGPT to research your product before they talk to anyone on your team — and not telling a soul? In this Research Radar Brief, Dr. Eva Wolf reviews 3 recent AI and marketing research papers covering token-level advertising inside AI-generated answers, autonomous AI-driven ad system optimization, and hidden consumer AI behavior in the healthcare journey. What you'll learn: - How a proposed auction system would let brands bid — word by word — to appear naturally inside AI-generated answers, not beside them - Why a purpose-trained AI model roughly doubled the success rate of senior human experts at improving an ad recommendation system - Why most survey respondents said they used ChatGPT to research health questions before a doctor visit — but didn't mention it to their physician - What the hidden AI research stage in the consumer journey means for health marketers right now - Why off-the-shelf tools like GPT-5.5 underperformed badly at specialized ad optimization, and what that suggests for how you build internal AI tools Papers covered: 1. Token-Level Advertising Authors: Hanbing Liu, Bowei Zhang, Changyuan Yu, Yinyu Ye, Qi Qi Source type: Preprint (not yet peer-reviewed) Access: Full text reviewed Source: https://arxiv.org/abs/2608.27382v1 2. Astar: Learning to Propose Evolution Directions for Self-Evolving Industrial AI Systems Authors: Jinxin Hu et al. Source type: Preprint (not yet peer-reviewed) Access: Full text reviewed Source: https://arxiv.org/abs/2608.27287v1 3. Generative AI use before medical visits: disclosure-item responses, trust, and care-seeking behaviors in a cross-sectional social-media survey in Poland Authors: Simona Wójcik, Anna Rulkiewicz, Justyna Domienik-Karłowicz Source type: Peer-reviewed journal article (Frontiers in Digital Health) Access: Full text reviewed DOI: 10.3389/fdgth.2026.1933451 Full show notes, transcript, and citations: https://bigplans.media/episodes/ai-token-advertising-self-evolving-ad-systems-health-ai-2026-08-28 Disclaimer: This is a first-pass research briefing produced by Evita, an AI-generated avatar trained on Dr. Eva Wolf's research framework. It is not a final academic review. Preprints have not been peer-reviewed and findings may change. All claims are attributed to the cited papers; listeners should consult the original sources before acting on any findings. -- 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 Wolf, marketing professor, AI researcher, and founder of Big Plans Media. Every day, Avita 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 ad slot you're bidding on right now is about to become irrelevant? Not because people stop clicking, because AI generates the answer before they ever see a result. And while that's happening, your customers are already using ChatGPT to research your product before they talk to anyone on your team, and they're not telling a soul. Today's papers point to the same pattern. AI is quietly restructuring the information layer between your brand and your buyer, and most marketing teams haven't noticed yet. We screened 382 papers. Three cleared the full text bar and made the radar. Quick caveat. This is a first pass research briefing, not a final academic review. 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. What does an ad actually look like inside an AI-generated answer? And who controls it? Because right now, if someone asks Chat GPT or Perplexity what running shoe to buy, there's no sponsored slot. There's just the answer. Brands have no formal way to influence it. This paper asks, can we change that? The researchers, this comes out of Stanford and Baidu, built something they call LAMA, token level advertising. And I want to make sure you actually hear this part. AI generates text one word at a time, token by token. This system lets advertisers bid on that process, not on a slot above the answer, not on a banner below it, on the words themselves, as they're being generated. Huh. So instead of winning a placement, you're winning influence over how the AI describes things. Your brand gets mentioned naturally inside the answer, not as an ad, as part of the response. And they built the auction so advertisers have an incentive to bid honestly. Gaming it doesn't help you. That's a mathematically provable property, which is honestly impressive. In simulation tests on real commercial search queries, Lama outperformed three comparison approaches on both revenue and answer quality. Now, here's the catch. It's a big one. This is a preprint marked work in progress, no real users, no live ad platform, no actual advertiser bids, theory plus simulation. The user perception question. Do people notice? Do they trust it? Do they object? Was not studied at all. And the disclosure implications? Completely unaddressed. Here's my honest reaction. This is the paper I'll be rereading in two years when a major AI search platform announces generation native ads. And I'll think, yeah, they read this too. That's the part I keep coming back to. Plain English payoff. The next version of a search ad isn't a slot you buy, it's a word you win inside the AI's answer itself. Okay, here's where this becomes commercially interesting. Money move. Start building what I'd call an AI response brief for your brand. How do you want an AI to naturally describe your product, in its own words, unprompted? That's what token level bidding would reward. Agencies that build this capability now own a new service line the moment any platform launches this format. Action step. Pull your top 10 paid search keywords today. For each one, ask, if an AI were answering this query naturally, how would I want my brand to appear in that answer? Write one sentence per keyword. That's your AI response brief, version one. Evidence check. Simulation only, no real users, unreviewed preprint. The findings are directionally compelling, but zero of this has been tested in the wild. Radar Verdict Watch list. The mechanism is genuinely novel and the team is credible, but it's not actionable until a real platform adopts it. Monitor closely. This next one looks like pure infrastructure. Stay with me because the business implication is bigger than it first appears. Paper two, here's the business question. Can AI replace the senior ML expert who decides what your ad system should test next? Because if you run a large ad platform, that person is expensive, rare, and a bottleneck. Every week they're deciding. What do we change in the model? What do we test? And they're often wrong. The researchers at Alibaba built a system called Astar. 8 billion parameters, trained not on the internet, trained on the history of their own ad system. Every model change, every experiment, every outcome. That's the training data. And here's what happened. Astar correctly suggested a useful improvement direction about 68% of the time. GPT 5.5, about 31%. Hmm, let me translate that. The purpose-trained AI was roughly twice as good at picking the right next step as either an expert human or the best general-purpose AI available. Not because it's smarter, because it learned from that specific system's history. They also ran Astar autonomously for two weeks on Lazada's live ad recommendation system. 20 improvement cycles, no human input, offline accuracy went up 23.6%, and in a live A-B test, real platform traffic, they saw 4.86% more total sales and 1.82% more ad revenue. Okay, here's the catch. This is one system, one company, one proprietary data set. To replicate this, you need hundreds of iteration commits with full code snapshots and training logs. Most companies don't have that. And this is still a preprint, not peer-reviewed. The GPT 5.5 result actually bothers me in a useful way. It tells you that off-the-shelf AI gives you generic advice that sounds smart and misses the mark every time. That's not a failure of GPT-5.5 generally. That's the wrong tool for the job. Plain English payoff. An AI trained on your own experiment history will dramatically outperform any general-purpose AI at improving your specific system. But first, you have to have that history. Here's the business hiding inside the research. Money move. If you're an ad tech consultant or ML infrastructure shop, there's a commit history intelligence service here. Go into large ad platform clients, clean and structure their iteration logs, build the training pipeline that makes this possible. Charge on a percentage of uplift. Action step. If you touch any ad system at all, start logging every model change and its outcome today. Structured, consistent, every single one. You're building the data set that makes this possible in two years. Future you will be grateful. Evidence check. Don't extrapolate widely. Radar verdict watch list. Directionally important, especially the start logging now implication, but not replicable for most teams today without Alibaba scale infrastructure. Paper three is the one I almost skipped. It's a healthcare study. But the consumer behavior signal inside it? Every marketer needs to hear this. Paper three. Here's the business question. Is AI already a hidden stage in your customer's decision journey and do they feel embarrassed about using it? Because that would change everything about where and how you show up. This is a peer-reviewed study out of Poland. Cross-sectional survey about a thousand adults recruited via Instagram. They asked people about using AI tools before medical appointments. Here's what they found. About 84% said they'd used AI to look up health information at least once. And among people who used AI before a visit, more than 80% did not tell their doctor. And not wanting to seem like they were challenging the doctor's expertise. So people are using AI. It's shaping what they think and what they ask. And the authority figure in the room has no idea. Now, let me be very clear about the limitations because they matter. Evidence check. The study couldn't confirm that people who said they used AI before a visit actually went to the doctor afterward. No validated scales, not pre-registered. The 82.5% non-disclosure figure is memorable. It is not a reliable rate. Do not cite it as fact. But here's why I'm not throwing this paper out. The directional signal is real. AI is a hidden pre-decision research step. People are embarrassed to admit it to authority figures. And fear of judgment is the number one reason they stay quiet. That is not a health care problem. That is a consumer behavior pattern. And I'm telling you, it exists in every high-stakes purchase category: finance, legal, big ticket retail, anything where admitting you Googled it feels like a weakness. If your customer is already using Chat GPT to research your product before they talk to a rep, a doctor, or an advisor, and they're not saying so, you're missing an entire stage of the journey, the AI stage. Plain English payoff. AI is already a hidden research step your customer takes before they ever talk to you, and they're embarrassed to admit it. Okay, here's where this becomes commercially interesting. Money move. If you're in health marketing, build a doctor-friendly AI summary feature. A one-tap printout of what the user researched via AI that they can bring to appointments. You close the stigma gap, you create a shareable artifact, and you own the AI research moment right before the clinical encounter. Action step. Audit your customer journey map before your next campaign review. Ask, is there an AI research step happening before they contact us? If yes, are we showing up in it? And does our messaging normalize that behavior or stigmatize it? Radar verdict. Use cautiously. The methodology is genuinely limited. Single country, convenient sample, self-reported. But the directional consumer behavior signal is worth taking seriously for any high-stakes category. Okay, let's put these three together. At first glance, these papers look separate, but together they show one thing. AI is restructuring the information layer between brands and buyers at every stage. And almost no one in marketing has built for it yet. Paper one says the ad slot itself is moving inside the generated answer, not above it. Paper two says the system deciding what your ads do is going to be run by a purpose-trained AI, not a senior human. Paper three says your customer is already in the AI research moment before they enter your funnel and they feel weird about it. The pattern, not more advertising, different architecture, not more personalization, different insertion points, not generic AI tools. AI that knows your specific system, your specific customer, your specific history. Here's the tension I keep coming back to. Papers one and two are about platforms and systems. Paper three is about people. And the people are ahead of the platforms. Customers are already in the AI research moment. The ad option for that moment doesn't exist yet. The system that could optimize for it requires data most companies haven't started collecting. That gap between where customer behavior already is and where marketing infrastructure is, that's the real finding today. Not hype, architecture. And the window to build it before someone else does is now. Here's the playbook from today. One, build your AI response brief. Take your top paid search keywords and write one sentence per keyword describing how you'd want an AI to naturally describe your brand in an answer. Do it before your next campaign review. Two, start logging every model or campaign change your team makes. Date, what changed, what the outcome was. Structured and consistent, starting today. You're building the data set that makes purpose-trained optimization possible. 3. Add an AI research stage to your customer journey map. Identify where customers are likely using Chat GPT before they contact you. Ask whether your content shows up there and whether your messaging makes them feel okay about it. Evidence check on all of that? Papers one and two are unreviewed preprints. Paper three is peer-reviewed but methodologically limited. Single country, convenient sample, self-reported. Use these findings 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.