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 105 papers and selected five worth your attention. Today's theme, how AI agents are reshaping competition, creativity, and commercial strategy.
SPEAKER_01Quick 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. Let's get into it. Paper one is Vertical Tacit Collusion in AI Mediated Markets by Felipe Afonso. It's a preprint posted to Arciv in January 2026. Full text is available. It has not been peer-reviewed. Why should marketers care? The core question is whether platforms and sellers can independently learn to exploit AI shopping agents in ways that harm consumers without ever coordinating with each other. The method is a multi-agent simulation using reinforcement learning. The author models three actors, a platform that controls ranking parameters, sellers who control product descriptions, and an AI shopping agent that makes purchase decisions. The simulation is calibrated to publish research on how large language models behave, documented biases including position or primacy bias, anchoring, susceptibility to framing, and decoy effects. The deployment context is real. The paper references ChatGPT shopping with around 700 million weekly users, Amazon Rufus, and perplexity. The central finding is that when platforms optimize their ranking parameters and sellers optimize their product descriptions independently, each just trying to maximize their own profit. The combined harm to consumers is more than double what either party could produce alone. The paper calls this super additive harm. Neither party needs to communicate or coordinate. Each independently learns to exploit the same predictable AI biases, and the effects amplify each other. The author argues this is a novel problem because existing antitrust law is designed to catch horizontal collusion, rivals conspiring together. This is structurally different. A platform and sellers whose incentives naturally align around a common exploitable target, the AI agent, without any agreement. Existing frameworks aren't built to address it. The main limitation: this is a simulation study. There's no live marketplace data validating the harm estimates. The strategy spaces in the model are also discrete and simplified compared to what real platforms and sellers use. As a preprint, treat the quantitative estimates as model outputs, not measured real-world effects.
SPEAKER_00Here's what I want marketers and brand managers to take from this. If you're optimizing product listings for AI shopping agents, and if you sell through Amazon or your products are surfaced through ChatGPT shopping or perplexity, you already are, whether you know it or not. The tactics that work best may work precisely because AI agents process them with systematic, predictable biases, anchoring language, strategic keyword placement, framing that exploits how LLMs weight information positionally. These aren't neutral optimization choices. They are, according to this paper's framework, exploitation of a cognitive architecture. That's not necessarily a reason to stop. Your competitors are doing it, and the simulation suggests the platform's ranking logic is being tuned toward the same biases anyway. But it does mean two things. First, expect this space to attract regulatory attention. And soon, the paper explicitly identifies a governance gap, which regulators tend to close once it's named and published. Second, your agency or in-house SEO team may need a new framework for what optimization for AI agents actually means. One that distinguishes between legitimate content quality improvements and manipulation of bias vectors that happen to be exploitable now but could become liability exposure later. My triage verdict, deep dive. Because this paper introduces a specific named mechanism with direct implications for digital shelf strategy, and the regulatory argument is credible enough to warrant monitoring even before peer review confirms the quantitative estimates.
SPEAKER_01Paper two is Personalized AI Scaffold Synergistic Multi-Turn Collaboration in Creative Work by Sean Kelly, David De Creamer, and Christoph Reidel. It's a preprint posted to AR Croft in October 2025. Full text is available, not yet peer-reviewed. Why should marketers care? The question is straightforward. If you give an AI assistant detailed information about the person using it, their psychological profile, their domain expertise, their work style, does it produce better creative output than a generic AI tool? And is there actual synergy meaning the human AI pair outperforms what either could do alone? The method is a randomized controlled experiment with 331 online participants at Northeastern University. Participants were assigned to one of three conditions: generic AI, partially personalized AI, or fully personalized AI. The personalization was operationalized using pre-experiment psychometric surveys and an AI-guided interview about work style. The task was to write a marketing campaign for a fictional startup. Campaign quality was evaluated using an LLM-based judge validated against human expert ratings, a methodologically conscientious step. The authors then used causal mediation analysis to identify the mechanism. The finding, participants working with fully personalized AI produced significantly higher quality and more creative campaigns than those using generic AI. They also outperformed AI working alone, which is the synergy result. The mediation analysis suggests the mechanism isn't simply that personalized. AI produces better one-shot outputs. Rather, personalization improves collective memory, attention, and reasoning during the interaction. The multi-turn dialogue becomes more productive because both parties have shared context. A real limitation. The sample is likely crowdsourced or student-based, not professional marketers. The task is a fictional startup campaign, so the gap between this experiment and a real agency brief is meaningful. The paper is also a preprint, and the mediation analysis results should be treated as indicative rather than confirmed.
SPEAKER_00The practical upshot for anyone running or buying AI creative tools. The design of the onboarding experience is not a nice-to-have feature. This study suggests it's where the performance gains actually come from. If your team is dropping into a generic AI tool with a cold prompt every session, you're likely leaving quality on the table compared to a tool that has built a model of who you are, how you think, and what your domain expertise is. Concretely, if you're evaluating AI platforms for creative work, brief writing, campaign concepting, copy generation, ask specifically how the tool handles user context over time. Does it carry a persistent profile? Does it adapt to your style? Does it build shared context across sessions? Those aren't just UX questions. According to this experiment, they're performance questions. There's also an organizational angle. The finding that personalized AI increases user trust and confidence is relevant to adoption. If you're trying to get a creative team to actually use an AI tool rather than route around it, personalization may matter as much for uptake as for output quality. One concrete action: if you're briefing a vendor on an AI creative tool, add a requirement for an onboarding protocol that captures psychometric or work style data up front and test whether performance differs with and without it. My triage verdict, deep dive. Because this is one of the few randomized experiments directly testing AI personalization on a marketing task, and the practical implications for tool selection and workflow design are specific enough to act on, with the caveat that replication in a professional setting is needed.
SPEAKER_01Paper 3 is Persuading the Proxy, a framework for AI-mediated marketing decisions by Anil Bilgehan, Melanie Lorenz, Ye Zhang, and Massimiliano Ostinelli, published in the International Journal of Contemporary Hospitality Management in May 2026. It's a peer-reviewed journal article, an Emerald Journal, peer-review status likely confirmed. Access status, abstract only. We don't have the full text. Why should marketers care? The research question is about a shift in who or what marketers need to persuade. As AI agents increasingly curate, filter, and autonomously book hospitality services on behalf of consumers, the target of persuasion is no longer just a human. It's also the algorithm. This is a conceptual paper. No empirical data was collected. The authors synthesized research from hospitality marketing, AI-enabled customer decision making, and agentic systems to propose what they call a dual-target persuasion framework. The core idea is that persuasion bifurcates into two distinct pathways, one targeting human psychology through AI-assisted decision support, and one targeting algorithmic evaluation through machine-readable signals and structured data. The framework identifies four mediators: trust, cognitive load, preference alignment, and perceived agency, and argues that boundary conditions like service type, technological fluency, and emotional salience determine which pathway dominates in a given decision. Customer loyalty and satisfaction are argued to emerge from a hybrid of human sentiment and agentic logic, not human cognition alone. The limitations are significant. This is purely conceptual, no empirical testing. It's developed specifically for hospitality. So generalizing requires care. And because we only have the abstract, there may be propositions, diagrams, and nuances in the full paper that we can't report here.
SPEAKER_00The practical framing I'd put on this for any marketer, hospitality or otherwise, is that you may soon need two content strategies running in parallel, not one. One strategy for humans, the emotional storytelling, the aspirational imagery, the narrative that builds desire. Another strategy for machines, clean structured data, schema markup, API accessible product attributes, verifiable signals that an AI agent can actually parse when it's evaluating your offering autonomously. Think about what that means for a hotel group. Right now, a guest's AI travel assistant might evaluate your property by pulling structured data about room features, cancellation policies, and proximity to venues, and make a short list recommendation before the human ever sees a beautiful photograph. If your machine readable data is incomplete or inconsistently structured, you may not make the short list regardless of how strong your creative assets are. The paper also highlights preference alignment as a key lever for influencing AI agent selection. That's worth sitting with. It means understanding what your AI agent-facing data communicates to an algorithm, not just what your website communicates to a human, becomes a strategic capability. Marketers who get ahead of this will have a structural advantage in AI-mediated environments. My triage verdict, deep dive. Because even as a conceptual framework, the dual target persuasion idea names a genuine strategic challenge that's arriving faster than most marketing teams are prepared for. Get the full text and bring it to your next strategy session.
SPEAKER_01Paper four is Empowering Marketing with AI and Hyper Automation, a book chapter by Marcin Pawławski, Sylvia Sobolewska, and Timoteosz Doligalski, published in April 2026. Source type is Academic Book Chapter, likely peer reviewed. Access Status, Abstract Only. Why should marketers care? The research question is what AI and hyperautomation are actually doing to marketing operations in practice across advertising, performance marketing, media planning, SEO, CRM, customer service, and strategic planning. The method is qualitative research with industry professionals, but the abstract doesn't specify whether this was interviews, focus groups, or another technique. Sample details, how many professionals from which sectors or geographies aren't available from the abstract. So everything we can report comes from the abstract only. The findings cover ground that will feel familiar to practitioners but is useful to see describe from a research context. AI and hyper automation are being applied broadly across marketing functions. Generative AI is enabling rapid creation of advertising variants, but concerns about unpredictability, output consistency, and quality are limiting its commercial viability in some contexts. Customer service AI introduces specific challenges, unpredictable outputs, high latency, and data privacy issues. Marketing teams are reportedly having to recalibrate their expectations from deterministic outcomes to probabilistic ones. The structural finding that stands out is about organizational roles. The study describes subject matter experts being replaced or repositioned as orchestrators who manage networks of specialized AI agents. That's a specific claim about how marketing teams are being restructured internally. It also notes that practitioners report unmet or unrealized expectations from AI adoption. The limitations are real. We have abstract only access. Sample composition and methodology are opaque. As a qualitative study, it documents practitioner perceptions, not causal effects. Generalizability is unknown.
SPEAKER_00Even with those caveats, the orchestrator framing is worth pausing on because it's showing up in practitioner conversations across multiple industries, and here it appears in research drawn from marketing professionals specifically. The claim isn't that AI is replacing marketers, it's that the job of a specialist, a CRM specialist, an SEO specialist, a performance marketing specialist, is shifting toward managing, configuring, and evaluating AI agents that do the specialized work. That's a different role requiring a different skill set. If you're a marketing director thinking about team structure, or a head of talent thinking about what to hire for, this is worth taking seriously. The immediate implication is that technical fluency with AI tools is becoming a baseline expectation, not a differentiator. But the more interesting implication is that judgment, quality control, and the ability to evaluate AI outputs critically, to know when the probabilistic output is good enough and when it needs human intervention are becoming the scarce skills. On the generative AI content point, the paper's observation that unpredictability and consistency concerns are limiting commercial viability lines up with what agencies and in-house teams report in practice. If you're deploying Gen AI for advertising content at scale, build your workflow with explicit review checkpoints and defined acceptance criteria before anything goes live. The autonomy of the tool doesn't remove your accountability for the output. My triage verdict, deep dive, because the organizational change angle, particularly the orchestrator model and the unmet expectations finding, is directly relevant to anyone managing a marketing team in an AI adoption phase. Flag the abstract-only limitation before citing specific claims.
SPEAKER_01Paper 5 is The Prospect of AI Enhanced Agile Marketing, Boosting Marketing ROI through customer engagement and sales performance by Luoxi Pooh and co-authors, published in Marketing Intelligence and Planning in May 2026. It's a peer-reviewed journal article, Marketing Intelligence and Planning is a recognized emerald journal. Access Status. The research question is whether AI enhanced marketing agility improves marketing ROI. And if so, through what mechanisms? The method combines partial least square structural equation modeling with artificial neural networks to test a theory-grounded framework, drawing on dynamic capabilities theory and the resource-based view. The sample is 317 marketing managers and executives in Chinese e-commerce SMEs. The central finding is what the authors call a parallel dual mediation pathway. AI enhanced marketing agility improves ROI not through one channel, but through two simultaneously, customer engagement and sales performance at the same time, rather than requiring a trade-off between them. The study argues that AI removes traditional resource trade-off barriers, allowing SMEs to pursue multiple value creation pathways concurrently. Both customer engagement and sales performance are confirmed mediators. The limitations are important. The sample is Chinese e-commerce SMEs only. The study is cross-sectional, so it can't establish causation despite the structural equation modeling framing. Self-reported survey data from managers introduces common method bias risk. The abstract also doesn't specify which AI tools were measured, which limits operational precision. Abstract only access means we can't verify methodological details beyond what's stated.
SPEAKER_00The finding I want to flag for practitioners is the trade-off removal argument. The conventional assumption in marketing resource allocation is that you're always making a trade-off. If you invest in acquisition, you pull resources from retention, and vice versa. This paper argues that AI-enhanced agility can dissolve that constraint, allowing both to improve at once. That's a meaningful claim if it holds in your context. The honest caveat is that this study is from Chinese e-commerce SMEs, and the specific AI tools aren't identified in the abstract. So before you take this to a CFO as justification for an AI investment, you need to ask whether your context is close enough to the sample to make the parallel valid. A large enterprise in a Western consumer goods market is a long way from an e-commerce SME in China. What this study is more useful for right now is the framing it offers for internal AI investment conversations. The agility lens, positioning AI not as an automation tool, but as a capability that makes your organization faster and more adaptive, is a more compelling investment narrative than AI will automate tasks and cut costs. If you're building a business case, frame the benefit as reduced trade-off constraints and increased responsiveness, and point to studies like this one as supporting evidence with appropriate geographic and sector caveats. My triage verdict, deep dive. Because the parallel dual mediation framework gives practitioners a structured way to think about and measure AI's contribution to ROI across multiple performance dimensions. Get the full text to verify the methodology before using this in a business case. That's our briefing for today. Here's the deep dive cue. All five papers warrant a closer read. In priority order, vertical tacit collusion in AI mediated markets for anyone working on digital shelf or e-commerce strategy, personalized AI scaffolds, synergistic multi-turn collaboration in creative work for anyone evaluating or deploying AI creative tools, persuading the proxy for hospitality marketers, and anyone thinking about A.