Exploring AI-Driven Product Discovery

We're excited to share an unexpected journey with you. What began as a simple question—"How do AI assistants recommend products?"—led us to insights that have fundamentally changed how we think about brand visibility and market research.

🔍 Our Experimental Approach

Honestly, we weren't sure what we'd find when we began this project. We only knew we wanted to do a hands-on, experimental study building on our Guide to AI Brand Visibility.

We selected "running shoes for beginners" as a test category—not because we're footwear experts but because it offered a defined product space with established brands and clear consumer needs (and because one of us plans to start running).

Our approach was straightforward:

• Query 10 leading AI models with 100 identical prompts, asking them for their top 5 product recommendations

• Collect 5,000 product recommendations in total, together with reasoning

• Analyze the patterns without preconceived expectations

What surprised us wasn't just the results but how powerfully this approach revealed category and competitive insights that traditional research methods might miss.

📊 The Process Became the Insight

We expected to learn about AI visibility patterns, but we discovered something more valuable: a research methodology that efficiently extracts category intelligence.

When multiple AI models independently highlight the same product attributes, these become strong indicators of a product’s position in the market. The consistency across different models revealed underlying patterns that validated and challenged our understanding of what matters to consumers.

This wasn't about gaming AI systems—it was about listening to what they collectively tell us about products, categories, and consumer expectations.

🏆 What We Found Along the Way

The specific results for running shoes were interesting, but the broader implications for marketers were fascinating:

AI Models as Market Intelligence Aggregators

Each AI model synthesizes vast amounts of public information—reviews, articles, specifications, expert opinions—providing a distilled view of how products are perceived across sources.

Visibility Variations Tell Stories

The differences in how brands appear across AI models often reveal underlying positioning strengths and weaknesses. Some brands show strong visibility in search-oriented systems but disappear in conversational assistants.

Category Norms Emerge Naturally

Without explicitly asking for them, we saw clear patterns in what AI models considered important for beginners, suggesting evolving category norms and expectations.

💡 A New Research Tool in Your Arsenal

What excites us most isn't that we uncovered a way to “hack” AI recommendations. Rather, we discovered a remarkably efficient method to understand product categories and competitive landscapes.

This approach offers several advantages:

• Surfaces insights without the biases of traditional survey methods

• Provides structured competitive benchmarking without manual comparison

• Reveals emerging patterns in how products are evaluated and discussed

• Offers a nearly real-time view of positioning and messaging effectiveness

🚀 Where We Go From Here

We're still exploring the potential of this methodology. Each category we analyze reveals new applications and insights. While brand visibility in AI systems is certainly important, we've come to value even more the intelligence these systems can provide about markets, competitors, and consumer preferences.

We'd love to hear how you might apply similar approaches in your industry. This is a collaborative journey of discovery, and we're learning alongside you.

📘 An Invitation to Join the Conversation

We'd welcome the conversation if you're interested in discussing this research approach and its applications to your specific challenges. This isn't about selling a finished product but exploring a new frontier together.

Simply reply to this email — we’d love to start a discussion.

At Horizon, the most valuable insights often come from asking simple questions and following the evidence wherever it leads. We're grateful to have you along on this journey of discovery.

— Torsten and Peter