Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

Analyzing the consumer interaction with generative AI-provided content in digital marketing

Jun Deng, Changsong Song — 2026 — Electronic Commerce Research

peer reviewed journal article  ·   ·  deep dive

Key findings

• Consumers could tell the difference between AI-made and human-made marketing content — they weren't fooled.
• Seeing AI-generated content did not meaningfully damage people's trust in the brand behind it.
• People's likelihood to buy a product was not significantly hurt by the fact that the content was made by AI.
• The study suggests AI-generated content can work in digital marketing without seriously undermining how authentic a brand feels to consumers — but the authors warn against using it carelessly.

Marketing implications

• You can use AI to write or generate marketing content without it being a brand-killer — most consumers in this study didn't lose trust just because content was AI-made. Try it on lower-stakes content like product descriptions or social captions.
• Don't assume consumers are fooled by AI content; many will recognize it. Be intentional about quality and authenticity rather than hoping no one notices.
• This is a green light to scale content production with AI tools, but the authors' caution against careless use still applies — don't let quality slip just because output is fast and cheap.

Paper B

Fine-Tuned LLM as a Complementary Predictor Improving Ads System

Hui Yang, Daiwei He, Kevin Jiang, Taejin Park et al. — 2026 — ArXiv.org (preprint)

preprint  ·   ·  read now

Key findings

• A fine-tuned AI language model can look at a user's past behavior on Pinterest — what brands they bought from, what categories they browsed, their demographics — and accurately predict which advertisers they are likely to buy from next. This is a new use of AI: not picking which ad to show at the last second, but predicting advertiser matches earlier in the process.
• Feeding these AI-generated advertiser predictions into Pinterest's ad system improved both steps of how ads get selected: the early step (which ads even get considered) and the later step (which ad wins and gets shown). Both improved, which is rare — usually optimizing one hurts the other.
• The system worked in real production at scale, producing measurable improvements in business metrics during live A/B tests, not just in lab settings. This is notable because most LLM-for-ads research has not been validated in real live systems.
• The key insight is treating the LLM as a helper predictor — not replacing the existing ad system, but feeding it better signals. This avoids the usual problems of LLMs being too slow or expensive for real-time ad decisions.

Marketing implications

• If you work at or with a large ad platform, push your engineering team to explore 'LLM as a signal generator' rather than 'LLM as a decision-maker' — this paper shows that predicting likely advertiser matches earlier in the pipeline (before the final ranking step) can improve results across the board without blowing up compute costs.
• If you're a performance marketer spending budget on Pinterest or similar platforms, know that these platforms are actively building AI systems that predict which advertisers you are likely to win before you even bid — which means brand-level behavioral history (past conversions, categories) matters more than ever for who gets shown to relevant users.
• For ad tech product builders: the specific recipe here — structured user history + fine-tuned LLM + inject predictions at retrieval and ranking — is described in enough detail to attempt a version of it. The paper's architecture (user selection → feature compilation → prompt-based prediction → downstream consumption) is a replicable blueprint.

Paper C

A Review on Generative AI in Digital Marketing: Transforming Customer Engagement and Firm Performance

Xu Youyu, Basheer Al-haimi — 2026 — International Journal of Academic Research in Economics and Management Sciences

peer reviewed journal article  ·   ·  test this week

Key findings

• AI tools that generate personalized ads and content — such as chatbots, AI copywriters, and recommendation engines — can make marketing messages feel more relevant to individual customers, which tends to increase engagement and loyalty compared to one-size-fits-all campaigns.
• Companies that use AI to automate routine marketing tasks (writing copy, answering customer questions, optimizing SEO) can free up their human teams to focus on strategy and creative decisions, making the overall operation faster and cheaper.
• Most companies are using AI in isolated pockets — one tool for email, another for ads — rather than as a unified strategy. This fragmented approach limits the competitive advantages they could get from AI.
• Real barriers to getting full value from AI in marketing include AI systems that can be biased or unfair, risks of becoming too dependent on a single AI vendor, data privacy problems, and the ongoing need for humans to check and approve what AI produces before it goes public.

Marketing implications

• If you are running AI tools only for one channel (e.g., just email or just ads), you are likely leaving value on the table. Look at where your team still does repetitive work manually — brief writing, image resizing, FAQ responses — and test an AI tool there this month.
• Before scaling any AI tool across your marketing, put a human review step in place. This paper's core warning is that AI outputs can carry bias or misrepresent your brand — having someone check before anything goes live is not optional, it is the minimum responsible practice.
• When evaluating AI vendors, ask how easy it would be to switch to a different provider. Vendor lock-in is flagged as a real business risk — build your stack so you are not trapped if a platform changes its pricing or terms.

Apple Podcasts  ·  Spotify  ·  Buzzsprout