Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

Efficient LLM-based Advertising via Model Compression and Parallel Verification

WenXin Dong, Chang Gao, Guanghui Yu, Xuewu Jiao et al. — 2026 — arXiv (Cornell University)

preprint  ·   ·  read now

Key findings

• The combined system made LLM-powered ad delivery run more than 1.8 times faster in real production — meaning the same server can handle more ad requests in the same amount of time.
• Inference speed improved by over 78% compared to the baseline, while quality remained competitive (the ads it picked or generated were still about as good as before compression).
• Shrinking the model's internal data representation (from 16-bit to 4-bit numbers) and pruning less important parts of the model reduced its memory footprint to roughly 30% of the original size, making it cheaper to host.
• A smart tree-based strategy that groups similar ad candidates and verifies many of them at once further cut the number of processing steps needed to generate each output.

Marketing implications

• If your ad platform uses an AI model to pick or write ads but it's too slow for real-time bidding, these compression techniques (quantization + pruning) are a proven path to cutting inference costs by half or more — worth raising with your ML engineering team.
• Ad agencies or in-house teams evaluating whether to run LLMs for real-time creative generation should know that speed barriers are solvable at production scale — the 'too slow and too expensive' objection is no longer a blocker for well-resourced teams.
• If you're building or buying an AI ad tool, ask vendors specifically about inference latency benchmarks and whether they use model compression — this paper shows those numbers should now be achievable well under 2× the cost of a simpler model.

Paper B

Generative AI in Marketing Communication: Consumer Perceptions, Brand Credibility, and Responsible AI Practices

Tahir Mushtaq, Onder Kethuda, Antje Cockrill, Ahmed Almoraish — 2026 — Cardiff Metropolitan Research Repository (Cardiff Metropolitan University) — AMI Conference 2026

conference paper  ·   ·  read now

Key findings

• Consumers who notice that a brand is using a lot of AI to create content start to trust that brand less — the communication feels less genuine and less unique.
• The researchers identified four specific ways this distrust shows up: (1) 'Verification burden' — consumers feel they have to fact-check AI content; (2) 'Emotional flattening' — AI content feels emotionally bland or hollow; (3) 'Content homogenisation' — everything looks and sounds the same because brands use similar AI tools; (4) 'AI fatigue' — people are simply tired of seeing AI-generated material.
• When a brand seems to rely too heavily on AI, it risks looking less credible and less distinctive than competitors who mix human creativity with AI tools.
• The authors argue that brands should use AI alongside human input rather than replacing human creativity entirely — this hybrid approach is positioned as the way to stay authentic and trustworthy.

Marketing implications

• If your brand publishes a lot of content — emails, social posts, product descriptions — make sure some of it has clear human touches: real opinions, specific stories, distinctive voice. Consumers are starting to notice when everything sounds the same.
• Before you ship a big AI-generated campaign, read it out loud and ask: does this actually sound like us, or does it sound like every other brand? If it's hard to tell, add a human edit pass.
• Consider being upfront about how you use AI in your content — hiding it entirely may backfire if consumers figure it out. Transparency about your human-AI mix could actually build trust rather than erode it.

Paper C

Sustainable Influencer Marketing: Leveraging AI and AR Tools to Promote Green Lifestyles

Lina Tio, Salamiah Muhd Kulal, Dorris Yadewani — 2026 — International Journal of Islamic Business and Management Review

peer reviewed journal article  ·   ·  use cautiously

Key findings

• When organizations use AI to pick influencers, they can find people whose audiences genuinely care about the environment — rather than just picking whoever has the most followers. This means the message reaches people more likely to act on it.
• Augmented reality (AR) features in campaigns — like letting users virtually 'see' a sustainable forest or product process — appear to make green messages feel more real and believable to audiences compared to standard photo or video posts.
• AI-generated visuals and AR simulations also carry a serious risk: they can make sustainability claims look credible even when the underlying facts are shaky. The paper warns this is essentially a high-tech greenwashing risk.
• When followers sense that an influencer's environmental claims don't match reality (e.g., they promote a brand with a poor environmental record), trust collapses quickly — and this effect is stronger for influencers whose followers feel a personal connection to them.

Marketing implications

• If you're picking influencers for a sustainability campaign, try using AI tools (like Heepsy, Upfluence, or similar platforms) that filter by audience values and environmental interest — not just follower count. You'll reach people who actually care about the topic.
• If you're running a green or sustainability campaign, be careful about using AI-generated imagery or AR experiences that make your product look more eco-friendly than it is. Audiences are increasingly suspicious, and if they catch a gap between your claims and reality, trust collapses fast.
• Before launching an AR or AI-powered sustainability campaign, build in a fact-checking step: can every environmental claim in your campaign be verified by a third party? If not, the tech makes the risk worse, not better.

Apple Podcasts  ·  Spotify  ·  Buzzsprout