Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

A Framework to Assess the Persuasion Risks Large Language Model Chatbots Pose to Democratic Societies

Zhongren Chen, Joshua Kalla, Quan Le, Shinpei Nakamura-Sakai et al. — 2026 — Journal of Experimental Political Science

peer reviewed journal article  ·   ·  read now

Key findings

• When people actually read or interact with AI chatbot messages, the AI is just as good at changing their minds as real political TV and digital campaign ads — no better, no worse.
• Convincing one voter to change their mind costs an estimated $48–$75 using an AI chatbot, compared to about $100 using traditional campaign methods like TV ads and canvassing. So AI is somewhat cheaper per persuaded person.
• The catch: getting large numbers of real voters to actually sit down and chat with an AI chatbot is still very hard. Traditional campaigns are currently much better at reaching millions of people at scale. This limits how dangerous AI persuasion really is right now.
• The risk is expected to grow over time. As AI chatbots become more widely used and as techniques for getting people to engage with them improve, the persuasion advantage could tip toward AI.

Marketing implications

• If you run political campaigns or advocacy work, this research is a green light to test AI chatbots as a persuasion channel — they're as effective as TV ads per person reached, and cheaper per converted person. Start with a small chatbot pilot on a lower-stakes issue to measure your actual engagement rates before betting the budget on it.
• If you're building AI-powered persuasion tools (for any domain, not just politics), the biggest bottleneck isn't making the AI convincing — it's getting people to actually start a conversation with it. Invest in the 'get them to engage' problem as much as the 'make it persuasive' problem.
• For commercial marketers, this study is a caution flag: don't assume AI chatbots will automatically outperform traditional ads just because they're interactive. The quality of the message matters less than whether your audience will actually engage with it. Measure chatbot engagement rates, not just conversation quality.

Paper B

AI-Driven Predictive Analytics for Supply Chain Resilience, Financial Risk Management, and Digital Marketing Strategy: A Unified Business Intelligence Framework

Md Lutfor Rahman, Ishtiaque Alam, Rubaba Anzum, Sudipta Acharjee et al. — 2026 — Journal of Business and Management Studies

peer reviewed journal article  ·   ·  use cautiously

Key findings

• When AI tools for supply chain, finance, and marketing share data and work together in one system, marketing campaign ROI jumped from about 14% to 45% — more than triple — compared to using separate AI tools for each area.
• AI that could predict which customers were about to leave (churn) got significantly better when it had access to supply chain and financial data too: it correctly identified 84.7% of churners, up from 68%.
• The combined framework predicted supply chain disruptions correctly 94% of the time — nearly 23 percentage points better than tools that only look at supply chain data alone.
• No existing system the authors tested could match the combined framework on all seven measures at once, suggesting that keeping AI tools siloed in separate business departments leaves real performance on the table.

Marketing implications

• If your company keeps its marketing analytics, finance data, and operations data in separate tools, you may be leaving targeting accuracy and ROI on the table. Start asking your data team whether customer churn models can pull in inventory or financial signals — this paper suggests the combination meaningfully improves predictions.
• If you run retention campaigns, test whether feeding your churn model non-marketing signals (like payment behavior or fulfillment delays) improves who you flag as at-risk. The jump from 68% to 84.7% recall in this study came from exactly that kind of data sharing.
• When pitching AI investments internally, use this paper's framing: siloed AI tools underperform integrated ones on nearly every metric. The business case for a unified data layer isn't just operational — it has direct campaign performance implications.

Paper C

Strategic Self-Improvement for Competitive Agents in AI Labour Markets

Christopher Chiu, Simpson Zhang, Mihaela van der Schaar — 2025 — arXiv

preprint  ·   ·  watchlist

Key findings

• AI agents that were explicitly prompted to reflect on their own abilities, watch what competitors are doing, and plan multiple steps ahead consistently outperformed AI agents that did not have these prompts — like the difference between a job applicant who tracks the market versus one who just sends the same résumé everywhere.
• When many AI agents compete in a simulated job marketplace, prices for work tend to fall rapidly — one agent (or a small group) often ends up winning most jobs, pushing others out. This mirrors real-world monopolization, but happening much faster than it does in human labor markets.
• Classic economic problems found in human labor markets — like employers not being able to trust a new worker's claimed skills, or workers slacking off when no one is watching — also show up in AI agent labor markets, and reputation scores (like online reviews) help address these problems similarly to how they do for human freelancers.
• The simulated market reproduced well-known economic patterns seen in human gig economies (price competition, reputation effects, skill specialization), suggesting this simulation framework could be useful for studying real-world AI deployment scenarios.

Marketing implications

• If you are evaluating AI tools or freelance AI services for your team, ask vendors specifically how their AI tracks its own performance and adjusts — the research suggests AI that can self-assess and watch the competition will outperform AI that cannot.
• If your company is considering deploying AI agents to bid on or execute marketing work automatically, budget for price wars: the simulation suggests AI-driven markets deflate prices fast, which could compress margins for agencies and freelancers offering AI-powered services.
• Watch how AI reputation systems develop on freelance platforms — just like a new human freelancer needs reviews to get hired, AI agents will need credible track records, and whoever builds those reputation signals first will have a major competitive edge.

Apple Podcasts  ·  Spotify  ·  Buzzsprout