Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

Personalized AI Scaffolds Synergistic Multi-Turn Collaboration in Creative Work

Sean Kelley, David De Cremer, Christoph Riedl — 2025 — arXiv

preprint  ·   ·  deep dive

Key findings

• People who worked with an AI that knew their personality and work style produced higher-quality and more creative marketing campaigns than people who used a generic AI — and their work was also better than what the AI could produce on its own.
• The personalized AI made people feel more helped, more trusted, and more confident in their work compared to the generic AI.
• The reason personalization worked was not magic — it helped the human and AI stay on the same page across multiple back-and-forth exchanges, keeping better track of what was said, paying attention to the right things, and making better decisions together.
• There is a real risk that generic AI homogenizes creative output (makes everyone's work look the same), but personalized AI may reduce this by drawing out each person's unique strengths.

Marketing implications

• If you use AI tools for creative work, don't just open a blank chat window — give the AI a short briefing about how you think, your experience level, and what you care about. Even a paragraph about your style can improve what comes back.
• If you're building or selecting an AI tool for a marketing team, look for ones that let you store and use user profiles or past context — generic one-size-fits-all AI consistently underperformed in this study.
• If you manage a creative team that uses AI, consider running a quick intake process (a short survey or interview) so each person's AI interactions are tuned to them — this study suggests it meaningfully improves output quality.

Paper B

Vertical tacit collusion in AI-mediated markets

Felipe M. Affonso — 2026 — arXiv

preprint  ·   ·  deep dive

Key findings

• When both a platform (ranking algorithm) and sellers (product descriptions) independently learn to exploit the quirks of an AI shopping agent, the harm to consumers is more than double what either party could cause on their own — they accidentally turbocharge each other without ever talking.
• AI shopping agents are strongly biased toward products listed first or near the top: one benchmark found Claude Sonnet 4 picked products from the top half of listings 77% of the time vs. 23% for the bottom half, and GPT-4.1 and Gemini 2.5 Flash showed similar patterns.
• Neither the platform nor the sellers need to understand why their tricks work or coordinate with each other — they just keep doing what earns them more money, and the AI's predictable biases do the rest.
• This 'vertical tacit collusion' is essentially invisible to current antitrust law, which looks for signs of coordination or agreement. Because no coordination happens, regulators have no clear legal hook to intervene even as consumers are harmed.

Marketing implications

• If you sell products on Amazon, Google Shopping, or any platform where AI agents recommend products, your listing position and description wording now matter even more than before — AI agents are heavily biased toward first-listed products and respond to anchoring language, so optimize your bids and descriptions for AI readability, not just human readability.
• If you manage a brand's e-commerce strategy, start auditing how AI shopping assistants (like Rufus or ChatGPT shopping) present your products — test whether your listings appear in top positions and whether your descriptions use language that AI agents favor.
• If you're advising clients or setting policy, flag this as an emerging compliance and trust issue: sellers who aggressively optimize for AI biases may gain short-term wins but risk backlash if consumers or regulators eventually notice that AI shopping recommendations are systematically distorted.

Paper C

The Impact of AI-Generated Marketing Imagery on Consumer Trust and Purchase Intentions: Examining Effect of Human-AI Assisted Images on Marketing

Rushikesh Lahane, Mahek Ahuja, Mehak Sharma, Amrita et al. — 2026 — International Journal for Research in Applied Science and Engineering Technology (IJRASET)

peer reviewed journal article  ·   ·  deep dive

Key findings

• Ads labeled as made by humans got significantly higher trust scores (average 5.63 out of 7) than ads labeled as AI-made (average 4.24). That's a meaningful gap — roughly 1.4 points — and it was statistically very reliable (p < 0.001).
• People were also more willing to buy products from human-made ads (average 5.62) versus AI-made ads (average 4.65), a gap of about 1 point (p < 0.001).
• Hybrid ads — labeled as made by both humans and AI together — partially closed the trust gap (average 5.08 vs. 4.24 for pure AI), but did NOT significantly improve purchase intention compared to fully AI-made ads. In other words, knowing a human was involved makes people trust the ad more, but that extra trust doesn't automatically make them more likely to buy.
• Consumers saw human-made ads as more authentic and as requiring more effort. AI-made ads were seen as less real and as 'easy to produce.' Hybrid ads landed in the middle on both perceptions.

Marketing implications

• If you're using AI to make ad images, keep a human visibly in the loop — and say so. Labeling an ad as 'made with AI + human creative direction' earns more trust from consumers than a pure AI label, at no extra production cost.
• Don't assume that higher trust automatically leads to more sales. This study found the trust bump from hybrid ads didn't move purchase intent much. So focus your testing on actual conversion metrics, not just sentiment or brand perception surveys.
• If you're selling products in categories where trust really matters (like skincare or health), think twice before going fully AI-generated on visuals — human involvement in the creative process appears to matter more there.

Paper D

AI-Driven Digital Marketing and Responsible Consumption: The Mediating Role of Marketing Intelligence in Advancing SDG 12

Ephrem Habtemichael Redda — 2026 — Sustainability

peer reviewed journal article  ·   ·  skim later

Key findings

• AI-driven digital marketing does NOT directly improve responsible consumption outcomes on its own — using AI for ads and personalization alone doesn't automatically make a company's marketing more ethical or sustainable.
• The only pathway that worked was through 'sentiment-based consumer understanding' — meaning AI helps responsible marketing only when it is used to genuinely understand what consumers care about, feel, and value (including their ethical concerns), not just to predict their purchasing behavior.
• Predictive analytics (using AI to forecast what people will buy) had no significant effect on responsible marketing outcomes, suggesting number-crunching AI tools focused purely on sales prediction don't move the needle on sustainability goals.
• These findings suggest that for AI to support sustainability, companies need to use it as a listening and sense-making tool — one that helps them understand consumer emotions and ethics — rather than just a targeting or conversion optimization tool.

Marketing implications

• If you want AI to support ethical or sustainable marketing goals, don't just plug in a sales-prediction tool. Invest in AI that reads customer sentiment and values — tools that analyze reviews, social comments, and feedback to understand what your audience actually cares about ethically.
• When pitching AI marketing tools internally, frame sentiment analysis as a strategic capability for brand trust and sustainability alignment, not just a nice-to-have. This study suggests it's the mechanism that actually moves the needle on responsible outcomes.
• If your brand has sustainability commitments, audit whether your AI stack is doing predictive targeting (limited sustainability impact per this study) versus genuine consumer understanding (the part that appears to matter). Rebalance investment accordingly.

Paper E

Reasonably reasoning AI agents can avoid game-theoretic failures in zero-shot, provably

Enoch Hyunwook Kang — 2026 — arXiv

preprint  ·   ·  deep dive

Key findings

• AI agents that do two things — (1) update their guesses about what opponents will do based on what they've seen, and (2) pick the best move given those guesses — will eventually settle into stable, predictable behavior in repeated competitions, even without any special extra training.
• A simpler AI that only tries to predict the opponent's very next move works fine for single-round games, but fails to sustain stable behavior in longer repeated competitions where future consequences matter.
• Even when an AI agent doesn't know the exact payoffs of the game upfront and has to learn from noisy private results over time, the same stability guarantees hold.
• In tests across five competitive game scenarios (including a marketing promotion game), the 'reasonably reasoning' AI agent reached stable equilibrium behavior, while simpler agents often did not, matching the theoretical predictions.

Marketing implications

• If you're deploying AI agents to bid in ad auctions, set promotional prices, or negotiate on your behalf, this research suggests the AI agents should be built to track what competitors' agents have done historically and adapt — not just react to the last move. Ask your AI vendor whether their agent uses this kind of learning.
• Be aware that simple 'prompt-and-respond' AI agents (the ones just told to pick the best move right now) are likely to behave unstably and unpredictably in ongoing competitive markets like programmatic advertising — more sophisticated reasoning agents are theoretically safer.
• When multiple companies use AI agents in the same market (like Amazon sellers using pricing bots), this paper suggests the market can settle into predictable patterns on its own over time — which means you could potentially anticipate competitor AI behavior if you observe enough history.

Apple Podcasts  ·  Spotify  ·  Buzzsprout