AI-Powered Shopping

Using technology to solve real user problems.

SUMMARY

Conceived and launched the first consumer-facing AI feature at CNN, Product Finder, which recommends the best products for users’ hyper-specific needs. Designed to help users quickly understand why a product fits their needs, these AI-generated summaries draw exclusively from CNN Underscored’s editorial testing—thousands of hours of hands-on reviews, expert analysis, and real-world product evaluations.

The feature translates Underscored’s proprietary product database into clear, contextualized, high-trust recommendations that both improve user satisfaction and support CNN’s commerce-driven revenue goals, resulting in +80% CTR and +48% AIV compared to organic affiliate content.

THE OPPORTUNITY

In partnership with the Machine Learning team, we originally launched Product Finder to deliver personalized product suggestions based on queries like “sunglasses for running” or “rain jackets that actually look cool.” However, users often needed to click deeper into articles to understand why a product matched their criteria, creating more work for them and ultimately friction in the conversion funnel.

CNN Underscored needed a way to surface this reasoning instantly—without compromising the brand’s journalistic standards or editorial voice. We needed to summarize thousands of hours of human-led testing for over 30k products within a reasonable time frame and budget. Enter: Generative AI.

USING DATA & ML WHILE STILL EMPHASIZING THE HUMAN

Our Commerce and Machine Learning teams collaborated to build an AI system capable of generating reliable, on-brand summaries grounded exclusively in Underscored’s editorial content. Any digital experience could scrape unreliable reviews and marketing pages for products — we needed to go a level deeper and show that expert human editors had their turn with the product to ensure it is worth getting behind.

Key components included:

  • Editorially-sourced training data to ensure accuracy and alignment with CNN standards

  • AI summaries written for clarity and intent—what the product is, why it fits the prompt, and how it performs

  • Rigorous review, QA, and iterative refinement to preserve trust and avoid hallucinations

  • Integration directly into Product Finder to streamline the path from query to recommendation to purchase

THE DRAMATIC RESULTS

Within the first month of launch, it was clear Product Finder was a win.

  • CTR (Click-through-rate): 80% higher than organic retail affiliate benchmarks. This reinforces the behavioral pattern: Product Finder attracts high-intent, end-goal users who are motivated to click off and purchase.

  • AIV (Average item value): 48% higher than organic retail affiliate benchmarks. Product Finder users demonstrate clearer goals and greater willingness to purchase higher-priced items, even without deals or discounts.

MORE THAN JUST A BUSINESS WIN

As the first generative AI feature at CNN, I and my team navigated the many technical and legal hurdles of using such a technology at a major media company focused on trust and user safety. Today, our launch serves as a blueprint for responsible AI product development across CNN’s experiences.

  • AI concept development, budget and feasibility analysis

  • Prompt engineering and model iteration

  • Editorial alignment, compliance, and style reinforcement

  • Scalable QA protocols and content safety frameworks

  • Cross-team workflow for future AI-powered consumer features

KUDOS

Product • Grace Hargis • Andrew Ross
Design • Agnes Kroczek
Engineering • Fahad Zafar • Ailish Burns
Editorial • Mike Bruno

And many, many more talented folks within and beyond Product Development that have made Underscored a success!