AI & Marketing Research Radar — 2026-05-06: AI-Generated Advertising and Consumer Trust

We screened 30 papers and selected 3. Theme: AI-generated advertising and consumer trust. Includes findings on disclosure granularity, AI avatar reviews, and consumer recognition of AI-generated ads.

Transcript

SPEAKER_03 0:00

Welcome to AI and Marketing Research Radar. I'm your host. Today we screened 20 papers and selected three worth your attention. Our theme, AI-generated advertising and consumer trust.

SPEAKER_01 0:11

Quick 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. Paper one is AI Generated versus Human Generated Advertising, Effects on Perceived Authenticity, Brand Trust, and Consumer Evaluation by Talon Manise, published in 2026. It's a university thesis from East Tennessee State University, published through their Digital Commons repository. It's likely peer-reviewed at the thesis level, but that's a different standard from a top-tier academic journal. Access status, abstract only. Our summary is based on the abstract alone. Full text, including sample size, stimuli details, and statistical results, was not available for review. Why should marketers care?

SPEAKER_00 1:12

The study asks a question: anyone running AI-generated creative right now is probably losing sleep over. Does telling people an ad was made by AI actually hurt the brand? The researcher ran a controlled experiment using Volvo and Jeep advertisements. Participants saw either AI-generated or human-generated versions, and some were told which type they were viewing. The study draws on priming theory and the elaboration likelihood model, a well-established framework for how people process persuasive messages. Because we only have the abstract. We don't know the exact sample size participant demographics, or the full statistical breakdown. That's a real constraint on how much confidence we can place in the specific numbers. Here's the core finding. When participants correctly identified an ad as AI generated, they rated it lower on authenticity, trustworthiness, and source credibility compared to human-generated ads. But here's the part that matters. No significant differences emerged on purchase intention, brand fit, or overall ad evaluation. AI disclosure alone, meaning just slapping a label on the ad, did not produce strong or consistent effects on perceptions. So the damage, when it happened, was upstream in credibility perceptions, not downstream in commercial outcomes. The key limitation, the effect only showed up when participants correctly identified the AI source. In the real world, most consumers aren't reliably doing that. We also have only two automotive brands and one thesis venue. So generalizability is genuinely uncertain.

SPEAKER_02 3:10

Here's what I take from this for a working brand team. The finding that purchase intention is unaffected even when credibility takes a hit is more reassuring than it sounds at first. But don't treat it as a green light to cut corners on AI creative quality. The mechanism matters. The credibility hit only landed when people detected the AI origin. So the strategic question isn't do we disclose, and it's how detectable is our AI creative and does it look like a machine made it? If your AI-generated video, copy, or imagery has that slightly uncanny quality that triggers recognition, you may be handing consumers a reason to discount the message even if they still click by out of habit. On disclosure, the finding that labels alone don't do much is important. If you're planning to add an AI-assisted tag as a trust management strategy, this study suggests it's probably not doing the protective work you think it is. A label without any accompanying signal of quality or human involvement may actually draw attention to the AI origin without providing reassurance. For automotive or any considered purchase category, I'd be especially careful. These are exactly the contexts where source credibility can eventually bleed into purchase hesitation, even if it doesn't show up cleanly in a single experiment. My triage verdict, deep dive, because it's directly on theme, the experimental design fits the question. And the split finding between credibility and commercial outcomes is precisely the kind of nuance that should shape how you brief your creative and disclosure teams.

SPEAKER_01 4:40

Paper two is biased by design by M. Jamon Jose and Jothus Rachel Matthews, published in April 2026. It's an academic book chapter in the advances in computational intelligence and robotics series, published by IGI Global, likely peer-reviewed at the book chapter level. Access status. Abstract only. Why should marketers care? The chapter asks whether AI systems used in neuromarketing, EEG measurement, eye tracking, facial coding, that kind of biometric consumer research, can pick up, amplify, or attenuate subconscious consumer biases. This is a conceptual paper. There's no primary empirical data. The authors apply dual process theory and algorithmic bias theory to build a proposed framework they call the ANBI model, AI Neuromarketing Bias Interaction, illustrated with case examples from retail, healthcare, entertainment, and political advertising. The models proposed, not empirically validated, their core argument. Combining AI with neurological consumer data enables highly targeted advertising and behavioral forecasting, but creates significant risks around consumer manipulation, privacy, and autonomy. They argue AI in this space should be oriented toward understanding consumers rather than exploiting them. And that transparency and ethical guardrails are required. Because this is abstract only, we can't assess the full argument, the depth of the case examples, or the specific guardrails they recommend. The primary limitation, this is theoretical. The framework is a lens for thinking, not a body of evidence. It draws on illustrative cases rather than systematic data, so its claims shouldn't be treated as proven.

SPEAKER_03 6:53

If your organization uses any form of AI-enhanced consumer research, biometric testing, attention analytics, emotion detection on ad content, this chapter flags a concern worth taking seriously, even without empirical proof. The argument is that the AI layer processing that biometric data may be trained in ways that introduce or amplify bias before the insight ever reaches your strategy deck. In practice, the objective neuromarketing data you're using to justify a creative decision or a targeting call might already be distorted by the system analyzing it. The actionable move, ask your neuromarketing vendor or tool provider one simple question. Has the system been audited for algorithmic bias and how it classifies emotional or attentional responses across different demographic groups? If the answer is vague or absent, that's a gap in your research quality assurance process. This matters most in sensitive categories, healthcare advertising, financial products, anything aimed at vulnerable populations. The ethical and reputational risk of AI systems that quietly exploit subconscious consumer responses isn't hypothetical, even if this paper can't quantify it yet. The NBI framework is a useful conceptual lens for anyone deploying AI and consumer research. But the absence of empirical data means you should treat it as a thinking tool, not evidence.

SPEAKER_01 8:20

Paper three is Atomic Fact Checking Increases Clinici in Large Language Model Recommendations for Oncology. Decision Support, a randomized controlled trial by Lisa C. Adams and colleagues, published in May 2026. It's a preprint on ARCUF and has not yet undergone peer review. Access status, full text available. Why should marketers care? This one needs a framing note up front. The study is conducted entirely in clinical oncology. It's not a marketing study, but the mechanism it tests is directly relevant to anyone building or buying AI-powered recommendation tools. And the effect size is large enough that it would be irresponsible not to flag it. The researchers ran a randomized controlled trial with 356 licensed clinicians, radiologists, radiation oncologists, and medical oncologists evaluating GPT 4.5 generated treatment recommendations across 21 clinical cases. The trial tested five transparency conditions: recommendation alone, recommendation with a natural language explanation, with source citations, with both, and with atomic fact-checking added on top. Atomic fact checking decomposes the AI output into individually verifiable claims, each linked to the specific source document supporting it. Trust was measured on a validated five-point scale, yielding 7,476 individual ratings. The design is consort compliant and IRB approved. The headline result, atomic fact checking, produced a Cohen's D of 0.94 on clinician trust versus controls, a large effect. The proportion of clinicians expressing trust rose from 26.9% to 66.5%, an increase of nearly 40 percentage points. Traditional approaches, explanation alone, citation alone, explanation plus citation, all improved trust over baseline. But their effects ranged from D of 0.25 to 0.50. Atomic fact checking nearly doubled the best conventional approach. The effect held across specialties, cancer types, and experience levels. Limitations, it's a preprint, not peer-reviewed. The cases were simulated, not real clinical decisions. The outcome is trust ratings, not actual clinical accuracy or patient outcomes. The sample is oncology clinicians, not consumers or marketers. And the translation to commercial AI tools requires conceptual work.

SPEAKER_03 11:18

The principle here is transferable even though the setting isn't. If you're presenting AI-generated recommendations to clients, creative recommendations, media mix suggestions, campaign performance forecasts, the instinct is usually to add an explanation or point to a data source. This study says that's not enough. The mechanism that actually moves trust is letting the person verify each discrete claim themselves, not handing them a paragraph that summarizes the AI's reasoning. In a marketing context, picture a platform that recommends increasing your video budget by 30%. A traditional explanation says video outperform display across your target audience last quarter. Atomic fact checking would decompose that into individual checkable claims. This specific placement delivered this CPM, this completion rate against this audience segment sourced from this campaign report dated this date, each claim independently checkable. That's a different interface design choice. And based on this study, it may make a material difference to whether your client or internal stakeholder acts on the recommendation or quietly files it away. If you're pitching AI generated strategy to a skeptical CMO, consider structuring your supporting materials so every factual claim is discreetly attributable and verifiable rather than woven into a narrative. My triage verdict. Skim later. The study is rigorous, and the mechanism is genuinely novel. But it requires significant conceptual translation before it's directly applicable to marketing practice. Worth keeping in your reading queue if you work on AI tool design, client-facing AI outputs, or internal AI adoption. That's our briefing. Here's the deep dive cue. Paper one AI generated versus human generated advertising by Talon Manise. The other two Biased by Design and the Atomic Fact Checking RCT land in the skim later queue for different reasons. The first for its conceptual framework on AI bias and neuromarketing, the second for its transferable trust design principle.

SPEAKER_01 13:32

As always, these are paper briefings, not final academic reviews. Paper one is a university thesis summarized from the abstract only. Paper two is a book chapter summarized from the abstract only. Paper three is a pre print and has not been peer reviewed. Links to all sources are in the show notes. Read the originals before making major decisions.

SPEAKER_03 13:58

See you next week.