Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

Assessing Capabilities of Large Language Models in Social Media Analytics: A Multi-task Quest

Ramtin Davoudi, Kartik Thakkar, Nazanin Donyapour, Tyler Derr et al. — 2026 — ArXiv.org

peer reviewed journal article  ·   ·  test this week

Key findings

• Modern AI models like GPT-4 can tell whether a given tweet was written by a specific person — but how well they do this depends heavily on which users and posts are chosen for the test. Using the most active users or topic-similar posts makes the task harder and exposes model weaknesses.
• AI models can generate tweets that somewhat match a person's style, but there is a gap between how good the tweets look to automated scoring tools versus how real users judge them — meaning a tweet can 'score well' on a computer metric but still feel fake to a human reader.
• AI models are better at guessing a user's broad interests (e.g., 'sports' or 'technology') from their tweets than at predicting their specific job title. Fine-grained occupation prediction remains a weak spot across all tested models.
• Newer, larger models (GPT-4, GPT-4o) generally outperform older or smaller ones (GPT-3.5, Llama 3.2) on these tasks, but no single model dominates across all three tasks — each has different strengths.

Marketing implications

• If you use AI to write social media posts in a brand voice or influencer's style, know that the posts may look good on paper but feel off to real followers — run a quick human review before publishing, not just an automated check.
• If you are building audience segmentation tools that infer user interests from social posts, AI models can handle broad categories (sports, tech, lifestyle) reasonably well — but don't trust them for fine-grained job-title targeting without human spot-checks.
• If you're worried about AI-generated content flooding your brand's comment sections or impersonating users, this research confirms that detecting fake posts is genuinely hard even for the best AI models — platform-level detection tools are not solved yet.

Paper B

Beyond the Creativity Paradox: A Theory-informed Framework for Role-based Integration of Generative AI in Organisational Creativity

Youngseok Choi, Chang Won Park, Ceyda Paydas Turan, Habin Lee — 2026 — Information Systems Frontiers

peer reviewed journal article  ·   ·  test this week

Key findings

• AI tools like ChatGPT or image generators are best understood as creative collaborators rather than just productivity shortcuts — they can generate idea variations, connect different knowledge areas, and speed up testing, while humans stay in charge of final decisions.
• The paper proposes four specific jobs AI can do in a creative team: (1) Creative Generator — brainstorming lots of ideas fast; (2) Conceptual Synthesiser — combining ideas from different fields; (3) Strategic Framer — helping shape and position concepts; (4) Human-Centred Facilitator — supporting the creative process itself.
• There is a 'Creativity Paradox': using AI heavily for creative work may quietly reduce people's personal motivation to create, make them less willing to take creative risks, and lead to blander, more predictable outputs — because the AI tends to produce what is statistically common, not genuinely novel.
• The authors argue that the risk is not that AI replaces human creativity, but that over-reliance on AI outputs can subtly shrink the range of ideas people explore, nudging teams toward safe, average results rather than genuinely original ones.

Marketing implications

• If your team uses AI to brainstorm campaign ideas, assign it a specific job rather than just asking it for 'ideas' — for example, use it as a bulk idea generator first, then have humans filter and take the boldest ones forward, not the safest ones.
• Watch out for your team's creative risk-taking shrinking over time as AI becomes the default first step. Periodically run brainstorms without AI to keep people's own creative muscles active.
• When briefing AI tools for creative work, explicitly push for unusual combinations or unexpected angles — AI naturally gravitates toward common, average outputs unless prompted to go further.

Paper C

AI-Driven Marketing Capabilities and International Competitiveness

Manoj Govindaraj, D. Jishnu, Jenifer Lawrence, Duggirala Aravind — 2026

academic book chapter  ·   ·  watchlist

Key findings

• Companies in emerging markets can use AI tools — like customer analytics and personalized marketing — to better spot what global customers want and react faster to market changes.
• The paper maps three ways AI helps firms compete globally: sensing (spotting opportunities and threats), seizing (acting on those opportunities quickly), and reconfiguring (reshaping the business to stay competitive).
• Companies that use AI to make real-time marketing decisions — rather than relying on slow, manual analysis — are proposed to perform better in exports and enter new markets more successfully.
• AI is suggested to give smaller or emerging-market firms a way to 'punch above their weight' globally by giving them the speed and insight that were previously only available to large multinationals.

Marketing implications

• If your company is trying to expand into new international markets, this paper suggests starting with AI-powered customer analytics to understand what local customers actually want before committing budget — rather than assuming your home-market playbook will work.
• Real-time decision-making tools (like AI-driven ad targeting or dynamic pricing) may help smaller brands move as fast as larger competitors when entering unfamiliar markets.
• For agencies or consultants working with emerging-market clients, framing AI adoption as a competitive tool for going global — not just an efficiency play — may be a more persuasive pitch.

Apple Podcasts  ·  Spotify  ·  Buzzsprout