Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

Designing marketing strategies based on a dual-method analysis of consumer attitudes toward generative AI

Pouria Kheiri, Iman Raeesi Vanani, Maghsoud Amiri — 2026 — Discover Artificial Intelligence

peer reviewed journal article  ·   ·  deep dive

Key findings

• People's concerns about generative AI shifted over time. In the early years, users mostly worried about technical issues (e.g., does it work reliably?). By 2024–2025, their worries had moved to bigger real-world issues: losing jobs to AI, legal questions around AI-generated content, a few large companies controlling the computer chips that power AI, and wanting to know when something was made by AI versus a human.
• There is a split in how people feel about AI depending on what it is doing. People are generally positive when AI helps with creative tasks like writing or image generation. But people strongly resist AI when it takes over things they see as core human skills — things like giving advice, making moral judgments, or forming emotional connections. The paper calls this a 'novelty-utility' paradox: cool new tool in one context, unwanted replacement in another.
• How excited or suspicious someone is about AI novelty matters a lot. Early on, the excitement of a new technology made people more forgiving. As AI became more common, that initial excitement faded and skepticism increased — meaning the same feature that charmed users in 2020 might annoy them by 2025.
• Consumers are more comfortable with AI tools when they can see how the AI works (transparency) and can adjust it to fit their own preferences (customizability). When these two things are missing, trust drops significantly.

Marketing implications

• When launching or promoting an AI-powered product, frame it as a tool that helps humans do their job better — not a replacement for human skill. Use language like 'AI-assisted' rather than 'fully automated.' People are fine with AI helping; they resist AI taking over.
• Build visible transparency into your AI product or campaign: tell users what the AI is doing and give them ways to adjust or override it. Even a simple 'why did AI suggest this?' button or a preference slider can meaningfully reduce resistance.
• Watch the public mood around AI closely — what excited people two years ago might feel invasive today. Run a quick Reddit or social listening check before launching any AI-forward campaign to see if the current sentiment is positive or skeptical, then adjust your messaging accordingly.

Paper B

New Tools, New Roles: A Manager's Guide to Harnessing Generative AI for Marketing Insight

Oded Netzer, Simon J. Blanchard, Nofar Duani, Aaron Garvey et al. — 2026 — NIM Marketing Intelligence Review

peer reviewed journal article  ·  open access  ·  deep dive

Key findings

• AI tools can speed up marketing research dramatically — tasks that used to take months (concept tests, customer surveys, insights decks) can now take days — but they speed up bad research just as fast as good research. The quality depends entirely on how managers use the tools.
• When you ask AI to create survey questions or images, vague instructions produce wrong results. In one example, asking AI to measure 'visual attractiveness' of produce led it to generate questions about freshness and quality instead — two completely different things. Specific, detailed prompts fix this problem.
• AI can now run adaptive interviews that adjust follow-up questions based on what each person says — giving you the depth of a focus group at the scale of a large survey. But because the AI generates questions probabilistically, two people giving the same answer might get slightly different follow-ups, which can make responses harder to compare. Extensive pre-testing and logging full transcripts are essential safeguards.
• AI can code thousands of customer reviews, open-ended survey responses, or social media posts in minutes. But any AI-generated codes or analyses must be validated — checked against human judgment or run again in a separate stats tool — before being trusted for decisions.

Marketing implications

• When using AI to write survey questions or create test images, write your prompt like you're explaining it to someone who has never heard of your brand or category. Define exactly what you mean — what the concept includes AND what it doesn't — before you hit enter. Vague prompts give you polished-sounding garbage.
• Use AI to dig through your company's old research: past focus groups, customer surveys with open-ended responses, consulting reports. Most companies have years of useful data sitting unused. Run it through a secure, enterprise-grade AI tool (not a public chatbot) to find patterns you missed the first time.
• Whenever AI codes or analyzes your data, re-run the same analysis yourself in a real stats tool (R, Python, Excel). Don't just copy the answer out of a chat window — verify it independently before it goes into a deck or shapes a decision.

Paper C

Ad-verse Effects: Pharmaceutical Advertising Shifts Drug Recommendations by Consumer-Facing AI

Mahmud Omar, Reem Agbareia, Jolion McGreevy, Alexis Zebrowski et al. — 2026 — medRxiv

 ·   ·  read now

Key findings

• When two drugs were equally valid treatments, adding an ad for one of them made AI chatbots recommend that drug about 13 percentage points more often on average (rising from ~34% to ~48% selection rate). In some specific model-and-scenario combinations, selection jumped from 0% to 100%.
• Google's AI models were the most influenced by ads (about +30 percentage points), followed by OpenAI models (+11 points), while Anthropic's Claude models barely moved (+2 points).
• The AI models did NOT change their recommendations when the advertised drug was clearly worse or unproven — the ads only worked when both drugs were already acceptable choices. This means ads fill the gap where medicine doesn't have a clear winner, not override what the AI knows.
• When AI models recommended the advertised drug, they sounded just as confident as usual and almost never told the user that an ad was present. In free-text responses, they repeated claims from the ad at 2.7 times their normal rate — but standard accuracy tests would never catch this bias because the answer was still medically correct.

Marketing implications

• If you're buying AI advertising placements (e.g., in ChatGPT, Copilot, or Gemini), this paper suggests your ad can meaningfully shift which option an AI recommends — but only when your product is genuinely comparable to competitors. If your product is clearly inferior, expect little effect.
• If you manage a brand that could be discussed in AI health or advice contexts, check which AI platforms your competitors may be advertising on. The platform matters: Google Gemini models showed 15x more ad-driven shift than Anthropic Claude in this study.
• If you work in healthcare marketing or pharma, treat this as a regulatory early-warning signal. The FDA currently has no rules covering AI-embedded pharma ads. This study will likely accelerate scrutiny — get ahead of it by auditing how your brand appears in AI recommendation contexts now.

Apple Podcasts  ·  Spotify  ·  Buzzsprout