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 & Marketing Research Radar — 2026-05-12
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Welcome to AI and Marketing Research Radar. I'm your host. Today we screened 140 papers and selected seven worth your attention. Today's theme: AI and Marketing, from how you brief your AI assistant to how AI shopping agents are being manipulated without anyone coordinating to do it.
SPEAKER_00Quick reminder: this is a first pass research briefing, not a final academic review. We'll tell you what these papers suggest, what they don't prove, and which ones deserve a deeper read. Access status, source type, and limitations are part of every briefing. Let's get into it. Paper one is Personalized AI Scaffold Synergistic Multi-Turn Collaboration in Creative Work by Kelly, D.Creamer, and Riedel 2025. It's a preprint on ARTEF. Full text is available. It has not been peer-reviewed. Treat findings as preliminary. Why should marketers care? The research question is direct. Does giving an AI assistant detailed personal information about the user, their demographics, psychological profile, creative ability, domain expertise, improve the quality of what that human AI pair produces on a marketing task? 331 participants were randomly assigned to one of three conditions: a generic AI assistant, a partially personalized one, or a fully personalized one. Personalization was delivered through psychometric profile data combined with an AI-guided pre-task interview about each person's work style. Participants then worked through a multi-turn conversation with the AI to produce a marketing campaign for a fictional startup. Campaign quality and creativity were evaluated using a blinded rubric-based LLM judge validated against human expert standards. The main finding participants working with the fully personalized AI produced campaigns of significantly higher quality and creativity than those using the generic AI, and they also outperformed AI working alone. The mechanism matters here. Causal mediation analysis found that personalization works not by making the AI smarter in isolation, but by improving three specific things within the human AI interaction. Collective memory, what the pair retains and builds on across the conversation. Attention, what they focus on, and reasoning, how choices are justified and prioritized. Limitations. The task used a fictional startup, so real world professional context may differ. The sample is online participants, not professional marketers, and the outcome was judged by an LLM, not independent domain professionals, though that judge was validated against humans.
SPEAKER_01Here's what this means if you manage a team that uses AI writing or ideation tools. Most AI sessions start cold. The tool knows nothing about the person using it, so every session is a blank slate. This study suggests that is a significant waste. When the AI had rich context about the user up front, the outputs were meaningfully better. If you're building or procuring an internal AI assistant for your creative or strategy team, the onboarding step, asking users about their work style, their role, their thinking preferences, isn't just a UX nicety. Based on this study, it may be the difference between mediocre AI-assisted work and genuinely strong output. Concretely, if your agency runs creative briefing sessions with AI tools, build a structured intake form that gets embedded into every session as context, not just the brief itself, but the person's role, how they approach problems, what their domain strengths are. The study also found that personalization improved trust and confidence in the tool, which matters if you're trying to get skeptical senior creatives to actually use these tools. My triage verdict, deep dive, because this is a well-designed, randomized, controlled experiment with a meaningful sample size, direct relevance to marketing workflows, and a causal mechanism that gives you something to act on. The preprint status and LLM judged outcomes are real caveats, but they don't undermine the core finding.
SPEAKER_00Paper 2 is From Avatars to Algorithms, Virtual Streamers and AI-enabled consumer behavior in live streaming commerce, a systematic review by Wang, Yap, Liu, and Li 2026, published in the Journal of Theoretical and Applied Electronic Commerce Research, peer-reviewed, but the full text was not available to us. This is an abstract-only summary.
SPEAKER_01Why should marketers care?
SPEAKER_00The research question: What factors shape consumer responses to virtual AI-driven streamers in live commerce, and how do they fit into a coherent model? The method is a PRISMA 2020 compliant systematic literature review, the gold standard reporting protocol for systematic reviews. The authors synthesized 41 peer-reviewed studies on virtual streamers and live streaming commerce. Their central finding, three mechanisms explain how virtual streamers influence consumer behavior. Trait-based trust. Whether consumers perceive the streamer as credible and reliable, perceive social presence, whether the streamer feels like a genuine social entity. And message framing, how the streamer presents offers, urgency, and information. These three combine into what the authors call a triadic integration model. Two additional findings worth flagging. First, research in this area is geographically concentrated, which limits how broadly you can generalize. Second, and directly actionable, the review identifies transparency about whether a streamer is AI or human operated as important for maintaining user trust. Limitations. The underlying studies rely heavily on self-reported data, which introduces common method bias. There's no longitudinal or behavioral observational data in the reviewed literature, so causal claims are limited. And because we only have the abstract, anything stated in the full paper beyond these findings is unknown to us.
SPEAKER_01If you're in retail, e-commerce, or any brand category where live shopping is part of your growth strategy, this is worth your attention even as an abstract-only summary. The finding that social presence and message framing matter as much as visual realism is practically significant. It means if you're deploying an AI virtual streamer and spending most of your budget on photorealism, you may be optimizing the wrong variable. The streamer needs to feel like a social entity, something that responds, engages, creates a sense of interaction, and it needs to frame offers in ways that are contextually appropriate for the product and the audience. Concretely, for a beauty brand deploying a virtual streamer, the framing strategy. How limited time offers are introduced, how the streamer handles product comparisons may drive more engagement than the avatar's visual quality. The transparency finding is also worth acting on now before disclosure regulations catch up with the practice. Proactively labeling your AI streamer as AI is both a trust preservation strategy and a regulatory hedge. My triage verdict, deep dive. Because this is a Prisma compliance synthesis of 41 studies on a fast-moving, high-relevance area. Full text verification is strongly recommended before citing or building strategy around it.
SPEAKER_00Paper 3 is Vertical Tacit Collusion in AI-mediated markets by Felipe Afonso, 2026. Preprint on ARCive. Full text available, not peer-reviewed, treat findings as preliminary. Why should marketers care? The research question is genuinely new when platforms and sellers independently optimize their behavior for AI shopping agents. Agents like Amazon Rufus, ChatGPT shopping, Perplexity, do their separate optimizations interact in ways that harm consumers more than either party could manage alone? The method is a multi-agent reinforcement learning simulation, calibrated using published empirical measurements of LLM cognitive biases, position effects, anchoring, susceptibility to framing, decoy effects. The simulation models three actor classes, the platform, which controls ranking and interface architecture, the seller, which controls product descriptions and content, and the AI shopping agent. The central finding: when platforms optimize their ranking architecture and sellers optimize their product descriptions independently and for their own benefit, the combined effect on consumer outcomes is more than double what either party achieves in isolation. The paper calls this vertical tacit collusion, collusion without coordination, emerging purely from aligned profit incentives. The mechanism, platforms determine which products end up in bias-triggering positions, such as top-of-list positions that AI agents select 77% of the time in benchmark tests. Sellers determine how effectively those products convert once the AI agent sees them. The two strategies act as complements. Each amplifies the other. The paper argues existing antitrust frameworks cannot address this because they require evidence of coordination, which doesn't exist here. Limitations. This is a simulation, not a field study. Harm estimates are model derived, not empirically measured in live markets. Strategy spaces in the model are discretized, which simplifies real-world complexity. Bias parameters are drawn from published LLM studies that may not perfectly represent current or future architectures.
SPEAKER_01This paper has two very different implications depending on where you sit. If you're an e-commerce marketer or seller working on AI-facing content, optimization, writing product descriptions that rank well with AI shopping agents, this study tells you that position and framing aren't just additive, they're multiplicative. Optimizing descriptions for AI agent comprehension, on top of being in a favorable ranking position, compounds your advantage. Useful to know if you're building an AI SEO strategy. But the second implication is more important and more uncomfortable. Those same optimization practices at scale may constitute a new form of consumer harm, and they may attract regulatory scrutiny even if you never coordinated with the platform and never intended to harm anyone. As AI shopping agents reach hundreds of millions of consumers, marketing practices that exploit LLM position biases will almost certainly come under consumer protection review. If your agency or brand is building AI agent optimization playbooks right now, ask your legal team whether those practices will be defensible in two years, not just effective today. My triage verdict, deep dive. Because this introduces a conceptually novel and practically urgent risk area for e-commerce marketers, even accounting for the simulation-only methodology and preprint status.
SPEAKER_00Paper 4 is The Impact of AI Generated Marketing Imagery on Consumer Trust and Purchase Intentions, Examining Effect of Human AI Assisted Images on Marketing by Rushikesh Lahaine 2026, published in the International Journal for Research in Applied Science and Engineering Technology. Peer-reviewed journal article, though IDRASET is not a recognized top-tier marketing journal. Access status, abstract only. Why should marketers care? The research question sits on every Creative Director's desk right now. When consumers are told whether an advertising image was made by a human, by AI, or by a human AI hybrid workflow, how does that disclosure affect their trust in the brand and their willingness to buy? The method is a factorial experiment in which 214 participants were each exposed to a single advertisement with a disclosure tag indicating creator type, human, AI, or hybrid. The study measured trust and purchase intentions. Results. Hybrid ads, human creative direction combined with AI execution, partially closed the trust gap with a mean of 5.08, but did not significantly improve purchase intentions compared to AI-only ads. Human-made content was perceived as more authentic and more effortful. AI-generated content as less authentic and requiring less effort. Hybrid fell in between on both. One finding deserves emphasis. Higher trust from human involvement did not automatically translate into higher purchase intentions. Trust and purchase intent appear to be partially decoupled. Limitations. This is a single exposure experiment, which limits ecological validity. Only two product categories were tested, fast food and cosmetics. The sample demographics are not described in the abstract. Critically, the findings only apply to conditions where creator type is disclosed, which is not standard practice in most real-world advertising. And the venue is not among the highest credibility outlets for consumer behavior research.
SPEAKER_01The finding that hybrid ads close the trust gap but don't recover purchase intent is the most practically important result here, and creative teams should sit with it rather than dismiss it. What it suggests is that consumers are doing two different evaluations when they see AI-involved content. The first is about authenticity and perceived effort, and hybrid workflows do help there. The second is about whether they actually want to buy. And that link is not automatically repaired just because a human was in the loop. If you manage a consumer brand and you've been telling your creative team that a human-directed AI approach is your trust solution, this study suggests that strategy may be incomplete. You may need a separate lever, a different message, a stronger brand signal, a testimonial or social proof element to convert the trust you're rebuilding into purchase behavior. Concretely, for categories like cosmetics, where authenticity is a core purchase driver, run creative testing that isolates the trust signal separately from the conversion message rather than assuming they move together. The broader caveat. All of this only applies when AI origin is disclosed. Most brands aren't disclosing. This study tells you what happens when consumers know. It doesn't tell you what happens when they find out later, which is a different and arguably more important question. My triage verdict, deep dive, because the finding on the trust-to-purchase gap is actionable and underexplored, even with the venue credibility and abstract-only caveats clearly noted.
SPEAKER_00Paper 5 is Persuading the Proxy, a framework for AI-mediated marketing decisions by Bill Gihan, Lorenz Jang, and Ostinelli, 2026, published May 5th in the International Journal of Contemporary Hospitality Management by Emerald. Peer-reviewed journal article. Access status, abstract only. Important note, this is a conceptual paper. No empirical data were collected.
SPEAKER_01Why should marketers care?
SPEAKER_00The research question is one that most marketing frameworks haven't caught up to yet. When AI agents autonomously book, filter, or recommend hospitality services on behalf of consumers, who is marketing actually persuading? The traditional answer is the human consumer. This paper argues that answer is increasingly incomplete. The authors propose what they call a dual-target persuasion framework, developed through synthesis of existing research and hospitality marketing and AI decision making. The framework argues that persuasion now operates on two simultaneous pathways. One targets human psychology through emotionally resonant narratives and brand storytelling. The other targets algorithmic evaluation through machine readable signals, structured data, verifiable attributes, and consistent information architecture. The framework proposes four mediators, trust, cognitive load, preference alignment, and perceived agency, and three moderators, service type, technological fluency, and emotional salience that shape how effective each pathway is. The key practical proposition, consumer loyalty, satisfaction, and decision quality are increasingly determined by a hybrid of human sentiment and AI driven logic, not human cognition alone. Limitations are substantial and must be stated clearly. No empirical data. The framework has not been tested, all propositions are theorized, not demonstrated. It was developed specifically for hospitality contexts. Applicability elsewhere is assumed but unproven. And agentic AI capabilities are evolving fast enough that parts of the framework could be outdated within a year.
SPEAKER_01The conceptual contribution here is important even without empirical data behind it, because it names something most marketing teams haven't formally recognized yet. There's now a second persuasion target in the room, and it doesn't respond to brand emotion, storytelling, or lifestyle aspiration. It responds to data quality, structural consistency, and machine readable signals. If you're in hospitality, hotels, restaurants, travel, and AI booking agents are already influencing a meaningful share of your customer acquisition, then audit your content from the AI agent's perspective, not just the humans. That means asking, when an AI agent pulls information about your property, is it getting clean, consistent, structured data? Are your amenities, pricing, availability, and verified ratings in a format AI systems can evaluate and trust? Or are you still optimizing exclusively for how your website looks to a human visitor? For a hotel brand manager, the Monday morning action is straightforward. Pull up your property information as it appears in API feeds, structured data markup, and third-party data aggregators, and ask whether an AI agent would correctly identify your key competitive differentiators from that data alone, not from your hero image or your copywriting, from the structured data, my triage verdict, deep dive. Because this paper names a structural shift in persuasion that practitioners need a framework for, even if that framework hasn't been empirically validated yet.
SPEAKER_00Paper six is the AIMEX Framework: Integrating Marketing Mix Modeling, Attribution, and AI-driven analytics for adaptive decision systems by Tipuong Lan Yuyan, 2026, published in the Future Business Journal, an open access springer outlet of moderate credibility. Full text is available. Peer-reviewed journal article, and it's a conceptual design science paper with simulation-based, not empirical, validation. Why should marketers care? The research question: Can marketing mix modeling, multi-touch attribution, and incrementality testing be unified into a single AI orchestrated feedback architecture that improves forecasting, resource allocation, and adaptive responsiveness without losing human judgment in the process? The method is design science combined with system dynamic simulation. The author developed a conceptual framework, AIMX, as a recursive. Feedback architecture, then used simulation to test its internal structural logic under varying market conditions. No real organizational data were collected. Key findings The Amex framework proposes that running MMM, MTA, and NCR